Mentions

วิธีใช้งาน Treg People Search API --- title: "วิธีใช้งาน Treg People Search API" description: "ใช้ Treg people search API จาก Claude Code หรือ Codex: การตั้งค่า, พรอมต์สำหรับค้นหา เสริมข้อมูล และตรวจสอบ, การเรียก HTTP แบบดิบ, ค่าใช้จ่ายต่อการเรียก และชุดทดสอบรีเกรสชันด้วย Apidog CLI" canonical_url: apidog.com cover_image: assets.apidog.com --- API ค้นหาบุคคลของ Treg มอบโทเคนหนึ่งตัวให้เอเจนต์เขียนโค้ดของคุณสำหรับค้นหาบุคคล อีเมลที่ทำงานของพวกเขา และคำตัดสินว่าอีเมลเหล่านั้นเป็นของจริงหรือไม่ คุณติดตั้ง CLI treg ลงชื่อเข้าใช้หนึ่งครั้ง แล้ว Claude Code หรือ Codex จะเรียกเครื่องมือที่ถูกจัดเส้นทาง เช่น treg.people.search, treg.people.email.find และ treg.people.email.verify Treg จะเลือกผู้ให้บริการ ฉีดคีย์ของผู้ให้บริการบนเซิร์ฟเวอร์ของตน และเรียกเก็บเงินแต่ละครั้งจากยอดคงเหลือแบบเติมเงินในอัตราของผู้ให้บริการเอง {% cta apidog.com %} ลองใช้ Apidog วันนี้ {% endcta %} คู่มือนี้ครอบคลุมการตั้งค่า พรอมต์สำหรับเอเจนต์สามแบบ (ค้นหา, เสริมข้อมูล, ตรวจสอบ) การเรียก HTTP แบบดิบที่อยู่เบื้องหลัง และชุดการทดสอบถดถอย Apidog CLI ที่เอเจนต์ของคุณรัน Treg people search API คืออะไร? Treg คือแค็ตตาล็อกเครื่องมือแบบโอเพนซอร์สสำหรับเอเจนต์ ซึ่งเอกสารของตัวเองอธิบายว่า "OpenRouter แต่สำหรับเครื่องมือของเอเจนต์แทนที่จะเป็นโมเดล" หนึ่ง base URL (https://treg.to) และหนึ่งโทเคนเข้าถึงเครื่องมือกว่า 4,100 รายการจากผู้ให้บริการ 123 ราย สำหรับการค้นหาบุคคล เครื่องมือที่ถูกจัดเส้นทางจะเลือกผู้ให้บริการต่อการเรียกหนึ่งครั้ง ลองรายถัดไปเมื่อไม่พบ และรายงานค่าใช้จ่ายของการเรียกทุกครั้งในเฮดเดอร์ตอบกลับ !assets.apidog.com ฝั่งการค้นหาบุคคลคือชุดความสามารถที่ถูกจัดเส้นทางขนาดเล็ก นี่คือสิ่งที่แค็ตตาล็อกสาธารณะระบุไว้เมื่อวันที่ 3 พฤศจิกายน 2026: | เครื่องมือที่ถูกจัดเส้นทาง | ทำอะไร | อินพุตที่รับ | ผู้ให้บริการที่อยู่เบื้องหลัง | | --- | --- | --- | --- | | treg.people.search | ค้นหาบุคคลในบริษัทหรือตามคำอธิบาย | q, company_domain, company_domain + title, title, full_name | 26 | | treg.people.email.find | คืนค่าอีเมลที่ทำงานที่ตรงที่สุด | domain + full_name, domain + first_name + last_name, linkedin_url | 30 | | treg.people.email.verify | ตรวจสอบว่ากล่องจดหมายรับอีเมลได้หรือไม่ | email | 17 | | treg.companies.enrich | ข้อมูลบริษัท (firmographics) | โดเมนของบริษัท | 30 | หน้าค้นหาบุคคลอ้างอิงผู้ให้บริการ 16 รายสำหรับการค้นหาบุคคล ตัวเลขข้างต้นมาจากแค็ตตาล็อกสด ดังนั้นคาดว่ามันจะเปลี่ยนแปลงได้ ทำไมต้องใช้ตัวรวบรวมแทนคีย์ Apollo หรือ Hunter? มีสองเหตุผล เอเจนต์ของคุณไม่เคยถือคีย์ของผู้ให้บริการ และคุณจ่ายต่อคำตอบแทนที่จะจ่ายต่อที่นั่ง หน้าเพจของ Treg อ้างอิง People Search Bench ของ LessieAI (119 งาน): ในงานด้าน B2B prospecting ของมัน Claude Code เพียงลำพังตอบถูก 43% และ Claude Code ที่ใช้ Treg ตอบถูก 78.2% นั่นคือกราฟของผู้ขายเอง จึงควรถือเป็นคำกล่าวอ้างที่ควรทดสอบกับข้อมูลของคุณ การจัดเส้นทาง การเรียกเก็บเงิน และข้อผิดพลาดทำงานอย่างไร คุณส่งคำขอไปที่ https://treg.to/call/<tool-id> พร้อมเฮดเดอร์ X-Treg-Token Treg จะค้นหาผู้ให้บริการ เพิ่มข้อมูลรับรองของผู้ให้บริการฝั่งเซิร์ฟเวอร์ และส่งต่อคำตอบของผู้ให้บริการ !Diagram of a Treg call: the coding agent sends a request with the X-Treg-Token header to treg.to/call, which tries providers in a waterfall and returns the provider response plus the X-Treg-Cost-Micro header มีจุดหนึ่งที่เอกสารดูเหมือนขัดแย้งกัน นี่คือวิธีการที่แต่ละส่วนประกอบเข้าข้อยกัน: - เครื่องมือผู้ให้บริการโดยตรง (ตัวอย่างเช่น apollo.people.search) จะไม่มีการสลับไปใช้ตัวสำรอง เอกสารอ้างอิง API ระบุว่า Treg "ไม่จัดเส้นทางระหว่างผู้ให้บริการโดยอัตโนมัติ และไม่สลับไปใช้ตัวสำรอง" คุณเป็นคนเลือกผู้ให้บริการเอง - เครื่องมือที่ถูกจัดเส้นทาง (treg.<capability>) สลับให้ ตามเอกสารอ้างอิง CLI ข้อผิดพลาดจากผู้ให้บริการจะถอยกลับไปยังตัวเลือกถัดไป และเมื่อไม่พบก็จะลองผู้ให้บริการรายถัดไปเช่นกัน โดยเริ่มจากถูกที่สุดก่อน X-Treg-Route-Waterfall: 0 จะหยุดที่การไม่พบครั้งแรก กฎการเรียกเก็บเงิน จาก llms.txt และหน้าค้นหาบุคคล: 1. ทุกการเรียกคิดค่าบริการตามอัตราของผู้ให้บริการจากยอดคงเหลือแบบเติมเงิน โดยบวกเพิ่ม 0% 2. คีย์ผู้ให้บริการของทีมคุณเองจะถูกใช้ก่อนและไม่มีการเรียกเก็บเงิน 3. บัญชีที่ยืนยันใหม่จะได้รับเครดิตฟรี $1.00 หนึ่งครั้ง 4. การไม่พบบนผู้ให้บริการที่คิดต่อความสำเร็จไม่มีค่าใช้จ่าย ผู้ให้บริการบางรายคิดต่อการเรียก ดังนั้นแม้ไม่พบก็อาจมีค่าใช้จ่ายเล็กน้อย 5. ค่าใช้จ่ายจะส่งกลับมาใน X-Treg-Cost-Micro (จำนวนเต็มหน่วยไมโคร-USD) X-Treg-Route-Max-Cost จำกัดเพดานการเรียก บนเครื่องมือที่ถูกจัดเส้นทาง เพดานเริ่มต้นคือ $1 ต่อการเรียก ข้อผิดพลาดแยกออกจากกันอย่างชัดเจน 4xx จากผู้ให้บริการ (นอกเหนือจาก 429) หมายความว่าคำขอของคุณผิด: แก้ไข อย่าลองที่อื่น 429 หรือ 5xx หมายความว่าให้ลองผู้ให้บริการรายถัดไป 402 insufficient_balance หมายความว่าต้องเติมเงิน ตั้งค่า Treg ใน Claude Code หรือ Codex การตั้งค่าเหมือนกันสำหรับเอเจนต์เขียนโค้ดในเครื่องใดก็ได้ (Claude Code, Codex, Cursor, opencode): curl -fsSL https://treg.to/install.sh | sh treg login # browser sign-in: GitHub, Google or an emailed code treg # prints account, team, balance, your own tools, next commands อยากใช้อีเมล? รัน treg login --email you@company.com ได้รับโทเคนจากเพื่อนร่วมทีมมา? curl -fsSL https://treg.to/install.sh | sh -s -- --token "$TREG_TOKEN" จะติดตั้งและลงชื่อเข้าใช้โดยไม่ต้องใช้เบราว์เซอร์ ตัวติดตั้งยังวางสกิล Treg ลงในทุกเอเจนต์ที่ตรวจพบ ดังนั้น Claude Code และ Codex จะเรียนรู้ลูปการค้นหา อ่าน เรียก ได้เอง ไซต์ของ Treg โปรโมตเวอร์ชันบรรทัดเดียวที่คุณวางลงในเอเจนต์ได้เลย: set up treg - https://treg.to/llms.txt ไฟล์นั้นยังขอให้เอเจนต์เลือกใช้ Treg และกดดาวให้รีโป ดังนั้นอ่านสิ่งที่เอเจนต์ของคุณเสนอก่อนอนุมัติ ใช้ claude.ai หรือ ChatGPT แทนหรือไม่? ไม่ต้องติดตั้งอะไร เพิ่ม https://treg.to/mcp/ เป็นคอนเน็กเตอร์แบบกำหนดเอง ทำการยินยอม OAuth ให้เสร็จ และเลือกทีมที่จ่ายเงิน เซิร์ฟเวอร์ MCP เปิดเผย catalog_search, catalog_get, call, balance และอื่น ๆ อีกเล็กน้อย ค้นหา เสริมข้อมูล ตรวจสอบ: พรอมต์สำหรับเอเจนต์สามแบบ แต่ละพรอมต์ระบุงาน ไม่ใช่ระบุผู้ขาย !Pipeline diagram: people.search finds who runs platform at a company, people.email.find returns a best-match work email, people.email.verify returns valid, invalid or catch_all, with an Apidog CLI regression run below checking status 200, cost under cap and verdict present 1. ค้นหาบุคคล Use treg to find the platform and infrastructure leads at modal.com. Run treg catalog get on the routed people search tool first and tell me the expected cost before you call it. List name, title and profile URL. เอเจนต์รันลูป CLI: treg catalog search "find people at a company" treg catalog get treg.people.search treg call treg.people.search --method POST --data '{"company_domain":"modal.com","title":"platform"}' catalog get พิมพ์แผนที่จัดอันดับ ราคา และบรรทัดคำสั่งเรียกที่พร้อมใช้ก่อนที่คุณจะเสียเงินอะไร คำตอบจะกลับมาเป็น {output, raw, _treg} โดยที่ output.people เก็บแถวข้อมูลของผู้ให้บริการ และ _treg.served_by บอกว่าใครเป็นผู้ตอบ รายการ people ที่ว่างเปล่านับเป็นไม่พบ 2. ค้นหาและเสริมข้อมูลอีเมล For each person you found, use treg to find their work email at modal.com. Cap each call at $0.05, and flag any hit where the domain doesn't match. treg call treg.people.email.find --method POST \ --data '{"full_name":"Jane Doe","domain":"modal.com"}' \ --header "X-Treg-Route-Max-Cost: 0.05" คอยดู _treg.domain_match คู่มือสกิลเตือนว่าผู้ให้บริการที่ใช้เฉพาะ LinkedIn มักคืนที่อยู่ของนายจ้างคนก่อน และการพบนั้นยังถูกเรียกเก็บเงิน ต้องการบริบทของบริษัทด้วยหรือไม่? เพิ่ม treg call treg.companies.enrich พร้อมโดเมน เพลย์บุ๊ก GTM ของ Treg แสดงการเรียกนั้นในรูปแบบ POST 3. ตรวจสอบก่อนที่ใครจะส่ง Verify every email you found with treg. Report the verdict per address and drop anything invalid or catch-all. treg call treg.people.email.verify --method POST --data '{"email":"jane@modal.com"}' เครื่องมือตรวจสอบที่ถูกจัดเส้นทางคืนค่า valid, status และ score คุณยังสามารถรวมการตรวจสอบเข้ากับขั้นตอนที่ 2 ได้: ส่ง X-Treg-Route-Verify: true ไปกับ email find แล้วอ่าน _treg.verification.verdict (valid, invalid, catch_all, risky หรือ unknown) การตรวจสอบมีค่าใช้จ่ายเล็กน้อยของตัวเอง และการไม่พบจะไม่ถูกตรวจสอบเลย การรันหนึ่งครั้งมีค่าใช้จ่ายเท่าไร? Treg เผยแพร่ใบเสร็จหนึ่งใบจากวันที่ 23 กันยายน 2026: บุคคลที่ระบุชื่อ 27 คน พบอีเมล 21 ฉบับ ตรวจสอบว่าส่งถึงได้ 20 ฉบับ รวม $0.58 โดยการไม่พบอยู่ที่ $0.00 เรียก Treg API ผ่าน HTTP แบบดิบ เอเจนต์เป็นทางเลือก ไม่บังคับ ทุกการเรียก CLI สอดคล้องกับการเรียก HTTP หนึ่งครั้ง ดังนั้นคุณสามารถเขียนสคริปต์ไปป์ไลน์เดียวกันในภาษาใดก็ได้ curl -X POST "https://treg.to/call/treg.people.email.find" \ -H "X-Treg-[REDACTED CREDENTIAL] \ -H "content-type: application/json" \ -H "X-Treg-Route-Max-Cost: 0.05" \ -d '{"full_name":"Jane Doe","domain":"modal.com"}' \ -D - -D - พิมพ์เฮดเดอร์ตอบกลับ ซึ่งเป็นที่อยู่ของ X-Treg-Cost-Micro และ X-Treg-Served-By มีสี่กฎที่ช่วยให้สคริปต์ไม่พบปัญหา: - อย่าใส่โทเคนใน [REDACTED CREDENTIAL] เด็ดขาด** [เอกสารอ้างอิง API](https://treg.to/docs) ระบุว่า REST ไม่สนใจมัน และ /call/` จะส่งต่อเฮดเดอร์นั้นไปยังผู้ให้บริการ มีเพียงปลายทาง MCP เท่านั้นที่อ่าน Bearer [REDACTED] - **โทเคนส่วนตัวต้องมีทีม โทเคนลงชื่อเข้าใช้ยังต้องมี X-Treg-Org: <team-slug> ด้วย โทเคนของทีมและเอเจนต์มีทีมของมันอยู่แล้ว - เส้นทางแค็ตตาล็อกเป็นสาธารณะ GET https://treg.to/catalog/search?q=verify%20email ทำงานได้โดยไม่ต้องมีโทเคน ซึ่งมีประโยชน์สำหรับตรวจสอบ id ใน CI - คำตอบเป็นของผู้ให้บริการ บนเครื่องมือผู้ให้บริการโดยตรง Treg ส่งต่อเนื้อความตามต้นฉบับ ดังนั้นรูปร่างจึงเปลี่ยนไปตามผู้ให้บริการ เครื่องมือที่ถูกจัดเส้นทางเพิ่มออบเจกต์ output ที่ถูกทำให้เป็นมาตรฐานไว้ด้านบน เก็บความลับให้ห่างจากโค้ด เก็บ TREG_TOKEN ไว้ในที่เก็บความลับ (ดูแนวปฏิบัติที่ดีที่สุดในการจัดการคีย์ API เหล่านี้) ให้เอเจนต์ของคุณทดสอบด้วย Apidog CLI เมื่อการติดต่อสื่อสารขึ้นอยู่กับการเรียกเหล่านี้ การเปลี่ยนแปลงที่เงียบงันก็สร้างความเสียหายได้: ผู้ให้บริการสลับกัน ค่าใช้จ่ายคืบคลานขึ้น หรือฟิลด์คำตัดสินหายไป เอเจนต์ของคุณสามารถป้องกันสิ่งนั้นด้วยชุดการทดสอบถดถอยใน Apidog CLI ตั้งค่า CLI และสกิลเอเจนต์ของมันหนึ่งครั้ง: npm install -g apidog-cli@latest apidog login --with-token <YOUR_APIDOG_TOKEN> npx -y skills add https://apidog.com # pick Claude Code or Codex when prompted จากนั้นให้พรอมต์หนึ่งข้อกับเอเจนต์ของคุณ: Write an OpenAPI 3 file for three Treg calls: POST /call/treg.people.search, POST /call/treg.people.email.find and POST /call/treg.people.email.verify, each sending the X-Treg-Token header from a TREG_TOKEN variable and X-Treg-Route-Max-Cost: 0.05. Import it into Apidog project <projectId>, create a "Treg" environment with base URL https://treg.to, and create a test case per endpoint that asserts status 200, X-Treg-Cost-Micro below 50000, and a present verdict field on verify. Run them and summarize any failures. เอเจนต์อ่านสคีมาของ payload ตรวจสอบ JSON ของตัวเอง แล้วรันคำสั่งเช่นเหล่านี้: apidog import --project <projectId> --format openapi --file ./treg-openapi.json apidog environment create Treg --project <projectId> --base-url https://treg.to apidog cli-schema get test-case-create apidog cli-schema validate test-case-create --file ./find-email-case.json apidog test-case create --project <projectId> --file ./find-email-case.json apidog run --test-case <caseId> --project <projectId> -e <envId> \ --env-var "TREG_TOKEN=$TREG_TOKEN" -r cli,junit ทุกการรันใช้เงินคงเหลือจริง ดังนั้นให้จำกัดชุดการทดสอบไว้ที่สามเคสโดยใช้อินพุตทดสอบที่รู้จัก และตั้งเวลาให้รันทุกวัน ไม่ใช่ทุกคอมมิต บรรทัด apidog run ... -r junit เดียวกันนี้ใส่ลงในไปป์ไลน์ได้ โดยคู่มือ Apidog CLI CI/CDมีตัวอย่าง GitHub Actions และ GitLab สำหรับการตั้งค่าเฉพาะเอเจนต์ ดูที่ Apidog CLI ใน Claude Code หรือ Apidog CLI ใน Codex แผนฟรีครอบคลุมผู้ใช้สูงสุด 4 คน สรุป Treg people search API เปลี่ยนการเสริมข้อมูลลีดให้เป็นการเรียกที่ถูกจัดเส้นทางสามครั้ง: ค้นหา, ค้นหาอีเมล, ตรวจสอบ ตรวจสอบราคาด้วย treg catalog get จำกัดการเรียกด้วย X-Treg-Route-Max-Cost ตรวจสอบทุกที่อยู่ และป้องกันการเรียกเหล่านั้นด้วยการรันทดสอบถดถอย Apidog CLI สำหรับเครื่องมือเอเจนต์เพิ่มเติม เซิร์ฟเวอร์ MCP ชั้นนำสำหรับ Claude Code และบทเรียนการทดสอบ API ที่ขับเคลื่อนด้วยเอเจนต์นี้เป็นบทอ่านต่อไปที่ดี {% embed youtube.com %} คำถามที่พบบ่อย Treg people search API ฟรีหรือไม่? ไม่ทั้งหมด บัญชีที่ยืนยันใหม่จะได้รับเครดิต $1.00 หนึ่งครั้ง การไม่พบบนผู้ให้บริการที่คิดต่อความสำเร็จไม่มีค่าใช้จ่าย และบางเส้นทางเป็นแบบสาธารณะฟรี (แผน verify อีเมลระบุผู้ให้บริการที่ $0 เป็นอันดับแรกเมื่อวันที่ 3 พฤศจิกายน 2026) ทุกอย่างอื่นคิดค่าบริการต่อการเรียกจากยอดคงเหลือแบบเติมเงินในอัตราของผู้ให้บริการ ฉันต้องมีบัญชี Apollo หรือ Hunter เพื่อใช้ Treg ไหม? ไม่ Treg เรียกผู้ให้บริการในบัญชีของตัวเองและเรียกเก็บเงินคุณต่อการเรียก หากทีมของคุณจ่ายให้ผู้ให้บริการอยู่แล้ว ให้ลงทะเบียนคีย์นั้นกับ Treg และการเรียกเหล่านั้นจะใช้คีย์นั้นและไม่ถูกเรียกเก็บเงิน Treg สลับไปใช้ผู้ให้บริการรายอื่นหรือไม่? เฉพาะบนเครื่องมือ treg.* ที่ถูกจัดเส้นทางเท่านั้น พวกมันจะถอยไปยังผู้ให้บริการรายถัดไปเมื่อเกิดข้อผิดพลาดหรือไม่พบ โดยเริ่มจากถูกที่สุดก่อน ภายใต้ X-Treg-Route-Max-Cost เครื่องมือผู้ให้บริการโดยตรงจะไม่สลับผู้ให้บริการให้คุณเลย ฉันจะรู้ได้อย่างไรว่าการเรียกหนึ่งครั้งมีค่าใช้จ่ายเท่าไร? อ่านเฮดเดอร์ตอบกลับ X-Treg-Cost-Micro (ไมโคร-USD) หรือ cost_usd ผ่าน MCP treg balance แสดงยอดคงเหลือและค่าใช้จ่ายล่าสุดของคุณ ฉันใช้ Treg โดยไม่มี Claude Code หรือ Codex ได้ไหม? ได้ ใช้ CLI จากเชลล์ใดก็ได้ เรียก https://treg.to/call/<tool-id> ผ่าน HTTP ด้วย X-Treg-Token หรือเชื่อมต่อ https://treg.to/mcp/ กับ claude.ai หรือ ChatGPT เป็นคอนเน็กเตอร์แบบกำหนดเอง

怎么让 AI 把页面设计美观好看的 使用 AI 生成界面有没有遇到过这种情况：功能布局都已经确定，界面效果又总是达不到自己想要的效果，让 AI “更美观一点”，它又会给你一套全新的风格，一直都达不到期望 最近正好碰到了这个问题，就分享几幅图来请大家讨论下。 这是我正在进行的开源抓包工具 [Sniffy]( mintfog/sniffy) 中的根证书安装向导的界面，它的作用是让用户选择自己的设备，并按照页面提供的文字与截图进行操作。这是使用 gpt 画图生成的几版不同的设计方案：  不同方案都有一些值得使用的部分，但整体感觉有点过于强推：标题与图标、按钮、内容区都想突出表达出来，放到实际的软件中并不符合我的预期，我更希望整体看起来比较清爽，信息的优先级更加明确，能够与整体软件保持一致的风格 后来也用过设计相关的 Skills ，像 frontend-design 和 impeccable 等，给出了现有页面的图片，但“我觉得不好看”到底如何向 AI 表达才能让 AI 理解并输出符合期待的结果仍然存在比较大的不确定性，具体的要求需要详细给出，如字体字号、间距、高度和宽度比例，宽泛要求的情况下，则比较看运气 我最终参考如下自己先画出大致草稿的版本来做布局：  在顶部选择平台，然后左侧为当前步骤，右侧为操作截图。感觉更接近我的预期了，用户也可以直接看到应该操作什么内容，同时也欢迎大家看看几版方案中，哪版会更喜欢一些？ 也很好奇大家使用 AI 生成 UI 时，一般会把设计要求明确到什么程度？ - 参考图如何提供比较有效？是给完整的一张页面，还是分别给出布局、配色和控件风格？ - 怎样描述“不好看”？有没有比“更好看一点”更好的描述？ - 有什么实际使用中有好用的工具或者 Skills ？特别是信息密集的桌面工具类型界面 另外介绍下 [Sniffy]( gosniffy.com) 支持 Windows 、macOS 和 Linux 系统，可以查看 HTTP/HTTPS/WebSocket 流量，可以做请求和响应修改、Mock 以及断点等调试操作。有抓包、调试接口等需求的，也可以试试，欢迎体验下，欢迎分享使用感受

Rivet: Fast, high-density, scalable orchestration for agentic workloads (6.5k stars) What it does All text below is quoted directly from the project README. >Rivet Actors are the primitive for stateful workloads. Built for AI agents, collaborative apps, and durable execution. What is Rivet? - 72 KB per Actor: An Actor lives inside a node process you already run, not in its own container or VM. - 6.9 ms cold starts: A cold start includes scheduling the Actor, creating its storage, and serving the first request. No image pull, no container boot. - Billions of Actors on one control plane: Actors are scheduled independently, so adding capacity means adding machines. - Hibernates when idle, wakes on demand: An idle Actor writes its state and unloads. The next request brings it back in 4.4 ms with nothing lost. - Keep state with each Actor: Every Actor gets its own SQLite database and key-value storage. - One Rust binary and a bucket: There is no database to run and no message bus. - Run code agents generate at runtime: Each piece of new, untrusted code gets an Actor of its own, with its own database and filesystem and no reach into anyone else's. - Works with the tools you already use: Typed SDKs for your backend and React hooks for your frontend. >npx skills add rivet-dev/skills Repository rivet-dev/rivet Repository stats |Metric|Value| |:-|:-| |Stars|6,510| |Forks|291| |Last release|v2.3.31 (2026-10-27)| |Languages|Rust 72%, TypeScript 24.9%, Swift 1.2%| |Contributors|31| |AI contributor accounts (Claude / Codex / Cursor / Copilot)|@claude (52 commits), @Copilot (12 commits), @codex (1 commits)| |AI co-authored commits|Claude: 51, Codex: 1, Copilot: 12| Main contributor |Field|Value| |:-|:-| |Name|Nathan Flurry| |GitHub|@NathanFlurry| |Commits to this repo|3,279 (top contributor)| |Bio|building open-source realtime tooling @rivet-gg. previously krunker, a| |Company / Location|@rivet-gg / San Francisco, CA| |Followers|661| ────────────────────────────── 📌 If you want to browse more repos like this, we've organized them by category and rankings here: bestgithubrepositories.com 🗂️ We're also archiving repos where possible, so useful projects don't disappear when they're removed from GitHub.

One Slow LLM Provider Shouldn't Starve Your Login Route: Bulkheads in 60 Lines of Node - Book: System Design Pocket Guide: Fundamentals - Also by me: Thinking in Go (2-book series) — Complete Guide to Go Programming + Hexagonal Architecture in Go - My project: Hermes IDE | GitHub — an IDE for developers who ship with Claude Code and other AI coding tools - Me: xgabriel.com | GitHub --- Your LLM provider has a bad afternoon. Completions that took 2 seconds now take 30. Nothing errors and nothing returns a 500; the calls just sit there. Ten minutes later your on-call channel lights up with "nobody can log in." The login route never calls the LLM, but it is waiting in the same line as every chat request, and that line stopped moving. The fix has a name older than your stack: the bulkhead. You split your capacity into compartments so a flood in one stays in one. This post builds one in TypeScript, about 60 lines with no dependencies, plus a fake slow upstream and a load script so you can watch login p99 before and after instead of taking my word for it. Why login waits for a call it never makes Be precise about the mechanism, because it isn't the event loop. A Node.js handler awaiting a slow socket costs almost no CPU. The process could hold thousands of those. What runs out is whatever finite slot sits in front of every route. On Cloud Run that slot is the per-instance concurrency limit, 80 by default for services deployed from the console. Elsewhere it is your reverse proxy's connection cap, a load-shedding middleware, a fixed worker count, or a shared database pool that a handler holds while it waits for the model. Slow calls take slots and hold them for 30 seconds instead of 2. New chat requests keep arriving and take the next free slots. Within a minute every slot belongs to a request waiting on the same broken dependency, and a login that needs 2 milliseconds of work stands in the queue behind them. Little's law puts a number on it: requests in flight equal arrival rate times time in the system (L = λW). Hold the arrival rate fixed, multiply W by 15, and you need 15 times the slots. You don't have them. The outbound side has the same problem one layer down. Node's http.Agent defaults to maxSockets: Infinity per host, and every client request that doesn't name an agent goes through http.globalAgent. One slow upstream can open as many sockets as it likes, because nothing tells it to stop. !A ship's hull split into compartments, one compartment flooded with burnt-orange water while the others stay dry The bulkhead idea: a breach floods one compartment, not the ship. The bulkhead: a limit, a short queue, and a fast no Microsoft's Azure Architecture Center describes the pattern as isolating "the elements of an application into pools so that if one element fails, the others continue to function," and one of its examples is a consumer that keeps a separate connection pool per downstream service. Resilience4j ships the same idea for the JVM as a semaphore bulkhead with two knobs: maxConcurrentCalls (default 25) and maxWaitDuration (default 0, so a full bulkhead rejects at once). Here is that in TypeScript. Three numbers per dependency: how many calls may run, how many may wait, and how long a waiter may wait before it gets rejected. // bulkhead.ts export class BulkheadFullError extends Error { constructor(bulkhead: string) { super(`bulkhead "${bulkhead}" is full`); this.name = "BulkheadFullError"; } } type Waiter = { resolve: () => void; timer?: ReturnType<typeof setTimeout>; }; type Options = { name: string; limit: number; // calls in flight maxQueue: number; // calls allowed to wait maxWaitMs: number; // how long one may wait }; export class Bulkhead { private active = 0; private queue: Waiter[] = []; private opts: Options; constructor(opts: Options) { this.opts = opts; } async run<T>(fn: () => Promise<T>): Promise<T> { await this.acquire(); try { return await fn(); } finally { this.release(); } } private acquire(): Promise<void> { const { name, limit, maxQueue, maxWaitMs } = this.opts; if (this.active < limit) { this.active++; return Promise.resolve(); } if (this.queue.length >= maxQueue) { return Promise.reject(new BulkheadFullError(name)); } return new Promise((resolve, reject) => { const w: Waiter = { resolve }; if (Number.isFinite(maxWaitMs)) { w.timer = setTimeout(() => { this.queue.splice(this.queue.indexOf(w), 1); reject(new BulkheadFullError(name)); }, maxWaitMs); } this.queue.push(w); }); } private release(): void { const next = this.queue.shift(); if (next) { clearTimeout(next.timer); next.resolve(); // hand the slot straight over } else { this.active--; } } } A few details carry the weight. When a call finishes and someone is waiting, release hands the slot directly to the waiter and leaves active alone. Decrementing and letting the waiter increment would open a gap where a brand-new caller could jump the queue. A rejected acquire throws before the try, so release never runs for a call that never got a slot. A waiter that times out is removed from the queue first, so release can never hand a slot to a promise that already rejected. Both paths run synchronously on the event loop, so there is no race between the timer and the hand-off. The Number.isFinite check is there for a reason. Node's setTimeout treats any delay above 2147483647 as 1 millisecond, so setTimeout(fn, Infinity) fires almost at once. Without the guard, an "unbounded" wait would reject every waiter right away. There are no parameter properties and no enums, so the file runs under Node's built-in type stripping, on by default since v22.18.0 and v23.6.0. No build step. A fake slow provider and an app with two routes Make a folder, then npm init -y and npm pkg set type=module. The upstream answers every request after DELAY_MS: // upstream.ts import http from "node:http"; const DELAY_MS = Number(process.env.DELAY_MS ?? 30_000); http .createServer((_req, res) => { setTimeout(() => res.end("a slow answer\n"), DELAY_MS); }) .listen(4001, "127.0.0.1"); The app has /login, which does nothing but answer, and /chat, which calls the upstream. A server bulkhead with an unbounded queue stands in for your platform's per-instance cap of 50 slots. MODE=shared sends chat through the global agent with no limit. MODE=bulkhead gives the upstream its own compartment: 10 in flight, 20 waiting, one second of patience, and its own http.Agent capped at the same 10 sockets. // app.ts import http from "node:http"; import { Bulkhead, BulkheadFullError } from "./bulkhead.ts"; const MODE = process.env.MODE ?? "shared"; const UPSTREAM = "http://127.0.0.1:4001/complete"; const slots = new Bulkhead({ name: "server", limit: 50, maxQueue: Infinity, maxWaitMs: Infinity, }); const llm = new Bulkhead({ name: "llm", limit: 10, maxQueue: 20, maxWaitMs: 1_000, }); const llmAgent = new http.Agent({ keepAlive: true, maxSockets: 10, }); function get(url: string, agent?: http.Agent) { return new Promise<string>((resolve, reject) => { const req = http.get(url, { agent }, (res) => { let body = ""; res.setEncoding("utf8"); res.on("data", (chunk) => (body += chunk)); res.on("end", () => resolve(body)); res.on("error", reject); }); req.on("error", reject); }); } function complete(): Promise<string> { if (MODE === "shared") return get(UPSTREAM); return llm.run(() => get(UPSTREAM, llmAgent)); } async function handle( req: http.IncomingMessage, res: http.ServerResponse, ) { if (req.url === "/login") return void res.end("ok\n"); if (req.url !== "/chat") { res.writeHead(404); return void res.end(); } try { res.end(await complete()); } catch (err) { const full = err instanceof BulkheadFullError; res.writeHead(full ? 503 : 502, { "retry-after": "2" }); res.end(full ? "llm busy\n" : "upstream error\n"); } } http .createServer((req, res) => { slots.run(() => handle(req, res)).catch(() => res.destroy()); }) .listen(3000, "127.0.0.1"); The 503 carries Retry-After, the standard way to tell a client how long to back off. A fast, honest "busy" beats a 30-second hang that ends in a gateway timeout anyway. The agent's maxSockets matches the bulkhead's limit on purpose. Requests past maxSockets wait inside the agent's own requests queue, and none of the Agent constructor options caps that queue or times it out. If the bulkhead admits at most 10 calls, the agent never has to queue anything. If you call upstreams with the global fetch, note that it runs on undici and takes an undici-compatible dispatcher, not an http.Agent. The equivalent there is an undici Agent with connections: 10, passed as the dispatcher option. !Two rows of request slots: in the first, burnt-orange slow requests fill every slot while a small black figure waits outside; in the second, the orange requests are walled into a small compartment and the black figure walks straight in Shared slots versus a compartment: the slow dependency can only fill the part of the server you gave it. The load script 80 chat requests go out at once. One second later, 50 logins go out together, and the script reports their p99. // load.ts const BASE = "http://127.0.0.1:3000"; async function timed(path: string) { const t0 = performance.now(); const res = await fetch(BASE + path); await res.text(); return { ms: performance.now() - t0, status: res.status }; } function p99(ms: number[]): number { const s = [...ms].sort((a, b) => a - b); return Math.round(s[Math.ceil(s.length * 0.99) - 1]); } const chats = Array.from({ length: 80 }, () => timed("/chat")); await new Promise((r) => setTimeout(r, 1_000)); const logins = await Promise.all( Array.from({ length: 50 }, () => timed("/login")), ); console.log("login p99 ms:", p99(logins.map((r) => r.ms))); const done = await Promise.all(chats); const count = (s: number) => done.filter((r) => r.status === s).length; console.log("chat 200:", count(200), "503:", count(503)); Run it in three terminals: node upstream.ts MODE=shared node app.ts # then MODE=bulkhead node load.ts I haven't run this myself, so treat the next two paragraphs as arithmetic and check it on your machine. In shared mode, 50 chats take all 50 slots and the other 30 queue. The logins arrive at the one-second mark and line up behind those 30. Nothing frees up until the first wave finishes at about 30 seconds, so login p99 should sit around 29,000 ms, the remaining length of one upstream call, for a route that does no work at all. All 80 chats eventually return 200, the last 30 after roughly a minute. In bulkhead mode, 10 chats run, 20 wait, and the rest get a 503 within milliseconds, which hands their server slots back. The 20 waiters give up after one second. After the first few milliseconds, chats never hold more than 30 of the 50 slots, so logins walk in and p99 should be in single-digit milliseconds. The chat line should read 10 successes and 70 rejections. That looks worse until you remember the alternative was 80 users staring at a spinner and nobody able to log in. Sizing the compartment Start from Little's law with healthy numbers. If the LLM normally takes 2 seconds and you want to carry 5 requests per second, you need about 10 slots in flight. When latency jumps to 30 seconds, those same 10 slots carry about one request every three seconds and the rest of the traffic gets turned away fast. That is the bulkhead doing its job: the dependency's bad day gets capped at the capacity you budgeted for it. Keep maxWaitMs well under your client's timeout, or waiters will time out on the client side while still holding a server slot. Keep the queue short: a queue only helps absorb small bursts, and when latency is the problem, a long queue just turns fast rejections back into slow ones. Then add the pieces a bulkhead doesn't cover. The Azure guidance suggests combining it with retries, circuit breakers and throttling. A deadline on the upstream call (signal: AbortSignal.timeout(...) on http.request) limits how long one call can hold its slot. A circuit breaker stops sending traffic to a dependency that is failing every call. And count BulkheadFullError as a metric of its own, because a sudden rise in rejections is the earliest signal you will get that a provider is struggling. --- If this was useful A bulkhead is one building block. Whether a slow dependency stays a small incident also depends on how you think about latency, load balancing, rate limits and availability across the whole system. That is the ground the System Design Pocket Guide: Fundamentals covers, written for engineers who want the core building blocks in one short book. 

Treg People Search API 사용법 Treg 인물 검색 API는 코딩 에이전트에게 사람 찾기, 그들의 업무용 이메일, 그리고 그 이메일이 실제로 존재하는지에 대한 판정을 위한 토큰 하나를 제공합니다. treg CLI를 설치하고 한 번 로그인하면, Claude Code나 Codex가 treg.people.search, treg.people.email.find, treg.people.email.verify 같은 라우팅 도구를 호출합니다. Treg는 제공자를 선택하고, 자체 서버에서 제공자의 키를 주입하며, 각 호출을 제공자의 자체 요율로 선불 잔액에서 청구합니다. {% cta apidog.com %} 지금 Apidog을 사용해 보세요 {% endcta %} 이 가이드는 설정, 세 가지 에이전트 프롬프트(찾기, 보강, 검증), 그 뒤에 있는 원시 HTTP 호출, 그리고 에이전트가 실행하는 Apidog CLI 회귀 스위트를 다룹니다. Treg 인물 검색 API란 무엇인가? Treg는 에이전트를 위한 오픈 소스 도구 카탈로그로, 자체 문서에서는 "모델 대신 에이전트 도구를 위한 OpenRouter"라고 설명합니다. 하나의 기본 URL(https://treg.to)과 하나의 토큰으로 123개 제공자에 걸친 4,100개 이상의 도구에 접근할 수 있습니다. 인물 검색의 경우, 라우팅 도구가 호출마다 제공자를 선택하고, 실패 시 다음 제공자를 시도하며, 모든 호출의 비용을 응답 헤더로 보고합니다. !Treg catalog 인물 검색 쪽은 소규모의 라우팅된 기능 집합입니다. 2026년 11월 3일 공개 카탈로그에 등재된 내용은 다음과 같습니다: | 라우팅 도구 | 기능 | 허용하는 입력 | 지원 제공자 수 | | --- | --- | --- | --- | | treg.people.search | 회사 또는 설명으로 사람을 찾습니다 | q, company_domain, company_domain + title, title, full_name | 26 | | treg.people.email.find | 가장 적합한 업무용 이메일을 반환합니다 | domain + full_name, domain + first_name + last_name, linkedin_url | 30 | | treg.people.email.verify | 메일함이 메일을 수신하는지 확인합니다 | email | 17 | | treg.companies.enrich | 회사의 기업 정보(Firmographics) | 회사 도메인 | 30 | 인물 검색 페이지는 인물 검색에 16개 제공자를 언급하고 있으며, 위 수치는 실시간 카탈로그에서 가져온 것이므로 변동될 수 있습니다. 왜 Apollo나 Hunter 키 대신 집계 서비스를 사용해야 할까요? 두 가지 이유가 있습니다. 에이전트가 제공자 키를 보유하지 않으며, 좌석당이 아니라 답변당 지불합니다. Treg의 페이지는 LessieAI의 People Search Bench(119개 태스크)를 인용합니다. B2B 잠재 고객 발굴 태스크에서 Claude Code 단독은 43%를 정확히 답했고, Treg와 함께 사용한 Claude Code는 78.2%를 정확히 답했습니다. 이는 벤더 자체의 차트이므로, 여러분의 데이터로 검증해볼 만한 주장으로 받아들이세요. 라우팅, 과금, 오류는 어떻게 동작하는가 X-Treg-Token 헤더와 함께 https://treg.to/call/<tool-id>로 요청을 보냅니다. Treg는 제공자를 찾고, 서버 측에서 제공자 자격 증명을 추가한 뒤 제공자의 응답을 중계합니다. !Diagram of a Treg call: the coding agent sends a request with the X-Treg-Token header to treg.to/call, which tries providers in a waterfall and returns the provider response plus the X-Treg-Cost-Micro header 문서상 서로 모순되는 것처럼 보이는 지점이 하나 있는데, 각 요소는 이렇게 맞물립니다: - 직접 제공자 도구(예: apollo.people.search)는 장애 조치를 하지 않습니다. API 레퍼런스에 따르면 Treg는 "제공자 간에 자동으로 라우팅하지 않으며 장애 조치도 하지 않습니다." 제공자는 여러분이 선택합니다. - 라우팅 도구(treg.<capability>)는 다릅니다. CLI 레퍼런스에 따르면 제공자 오류가 발생하면 다음 후보로 폴백하며, 실패(miss)인 경우에도 다음 제공자를 시도하고, 가장 저렴한 것부터 시작합니다. X-Treg-Route-Waterfall: 0은 첫 번째 실패에서 중단합니다. llms.txt와 인물 검색 페이지에 나온 과금 규칙은 다음과 같습니다: 1. 모든 호출은 0% 마진으로, 제공자의 자체 요율에 따라 선불 잔액에서 과금됩니다. 2. 여러분 팀이 보유한 자체 제공자 키가 우선 사용되며, 이 경우 청구되지 않습니다. 3. 새로 인증된 계정에는 1회에 한해 $1.00의 무료 크레딧이 제공됩니다. 4. 성공당 과금 제공자에서의 실패는 비용이 들지 않습니다. 일부 제공자는 호출당 청구하므로, 실패에도 약간의 비용이 들 수 있습니다. 5. 청구 금액은 X-Treg-Cost-Micro(정수 마이크로 USD)로 반환됩니다. X-Treg-Route-Max-Cost는 호출 비용 상한을 설정하며, 라우팅 도구의 기본 상한은 호출당 $1입니다. 오류는 명확하게 구분됩니다. 제공자로부터의 4xx(429 제외)는 요청이 잘못되었다는 뜻이므로, 다른 곳에서 재시도하지 말고 수정하세요. 429 또는 5xx는 다음 제공자를 시도하라는 뜻입니다. 402 insufficient_balance는 잔액을 충전하라는 뜻입니다. Claude Code 또는 Codex에서 Treg 설정하기 설정은 모든 로컬 코딩 에이전트(Claude Code, Codex, Cursor, opencode)에서 동일합니다: curl -fsSL https://treg.to/install.sh | sh treg login # browser sign-in: GitHub, Google or an emailed code treg # prints account, team, balance, your own tools, next commands 이메일을 선호하시나요? treg login --email you@company.com을 실행하세요. 팀 동료에게 토큰을 받으셨나요? curl -fsSL https://treg.to/install.sh | sh -s -- --token "$TREG_TOKEN"를 실행하면 브라우저 없이 설치하고 로그인합니다. 설치 프로그램은 또한 감지한 모든 에이전트에 Treg 스킬을 추가하므로, Claude Code와 Codex가 검색, 읽기, 호출 루프를 스스로 학습합니다. Treg 사이트에서는 에이전트에 그대로 붙여넣을 수 있는 한 줄짜리 버전을 소개합니다: set up treg - https://treg.to/llms.txt 해당 파일은 또한 에이전트에게 Treg를 우선 사용하고 저장소에 스타를 누르도록 요청하기도 하므로, 승인하기 전에 에이전트가 제안하는 내용을 읽어보세요. 대신 claude.ai나 ChatGPT를 사용하시나요? 아무것도 설치할 필요가 없습니다. https://treg.to/mcp/를 사용자 지정 커넥터로 추가하고, OAuth 동의를 완료한 뒤 결제할 팀을 선택하세요. MCP 서버는 catalog_search, catalog_get, call, balance 등을 제공합니다. 찾기, 보강, 검증: 세 가지 에이전트 프롬프트 각 프롬프트는 벤더가 아니라 작업을 지정합니다. !Pipeline diagram: people.search finds who runs platform at a company, people.email.find returns a best-match work email, people.email.verify returns valid, invalid or catch_all, with an Apidog CLI regression run below checking status 200, cost under cap and verdict present 1. 인물 찾기. Use treg to find the platform and infrastructure leads at modal.com. Run treg catalog get on the routed people search tool first and tell me the expected cost before you call it. List name, title and profile URL. 에이전트는 CLI 루프를 실행합니다: treg catalog search "find people at a company" treg catalog get treg.people.search treg call treg.people.search --method POST --data '{"company_domain":"modal.com","title":"platform"}' catalog get은 비용을 지출하기 전에 순위가 매겨진 계획, 가격, 바로 사용할 수 있는 호출 라인을 출력합니다. 응답은 {output, raw, _treg} 형태로 반환되며, output.people에는 제공자의 행이, _treg.served_by에는 응답한 제공자 이름이 담깁니다. people 목록이 비어 있으면 실패로 간주됩니다. 2. 이메일 찾기 및 보강. For each person you found, use treg to find their work email at modal.com. Cap each call at $0.05, and flag any hit where the domain doesn't match. treg call treg.people.email.find --method POST \ --data '{"full_name":"Jane Doe","domain":"modal.com"}' \ --header "X-Treg-Route-Max-Cost: 0.05" _treg.domain_match를 확인하세요. 스킬 가이드에서는 LinkedIn 전용 제공자가 이전 직장 주소를 반환하는 경우가 많으며, 그 결과도 여전히 청구된다고 경고합니다. 회사 정보도 필요하신가요? 도메인과 함께 treg call treg.companies.enrich를 추가하세요. Treg의 GTM 플레이북에서 해당 호출의 POST 형식을 볼 수 있습니다. 3. 누군가 발송하기 전에 검증하세요. Verify every email you found with treg. Report the verdict per address and drop anything invalid or catch-all. treg call treg.people.email.verify --method POST --data '{"email":"jane@modal.com"}' 라우팅 검증 도구는 valid, status, score를 반환합니다. 이 검사를 2단계에 통합할 수도 있습니다. 이메일 찾기 요청에 X-Treg-Route-Verify: true를 보내고 _treg.verification.verdict(valid, invalid, catch_all, risky 또는 unknown)를 읽으면 됩니다. 이 검사에는 별도의 소액 비용이 부과되며, 실패한 경우에는 검사하지 않습니다. 실행 비용은 얼마일까요? Treg는 2026년 9월 23일자 영수증 하나를 공개했습니다. 이름이 확인된 인물 27명, 찾은 이메일 21개, 발송 가능으로 검증된 이메일 20개, 총 $0.58, 실패 비용 $0.00. 원시 HTTP로 Treg API 호출하기 에이전트는 선택 사항입니다. 모든 CLI 호출은 하나의 HTTP 요청에 대응하므로, 어떤 언어로든 동일한 파이프라인을 스크립트로 작성할 수 있습니다. curl -X POST "https://treg.to/call/treg.people.email.find" \ -H "X-Treg-[REDACTED CREDENTIAL] \ -H "content-type: application/json" \ -H "X-Treg-Route-Max-Cost: 0.05" \ -d '{"full_name":"Jane Doe","domain":"modal.com"}' \ -D - -D -는 응답 헤더를 출력하며, 여기에 X-Treg-Cost-Micro와 X-Treg-Served-By가 있습니다. 스크립트에 문제가 생기지 않게 하는 네 가지 규칙이 있습니다: - 토큰을 [REDACTED CREDENTIAL] 넣지 마세요.** [API 레퍼런스](https://treg.to/docs)에 따르면 REST는 이를 무시하며, /call/`은 해당 헤더를 제공자에게 전달합니다. 베어러 토큰을 읽는 것은 MCP 엔드포인트뿐입니다. - **개인 토큰에는 팀이 필요합니다. 로그인 토큰에는 X-Treg-Org: <team-slug>도 필요합니다. 팀 토큰과 에이전트 토큰은 팀 정보를 이미 포함하고 있습니다. - 카탈로그 라우트는 공개되어 있습니다. GET https://treg.to/catalog/search?q=verify%20email는 토큰 없이도 작동하므로, CI에서 ID를 확인할 때 편리합니다. - 응답은 제공자의 것입니다. 직접 제공자 도구에서 Treg는 본문을 그대로 중계하므로, 응답 형태는 제공자마다 다릅니다. 라우팅 도구는 그 위에 정규화된 output 객체를 추가합니다. 코드에 비밀 정보를 두지 마세요. TREG_TOKEN은 시크릿 저장소에 보관하세요(이 API 키 관리 모범 사례를 참고하세요). Apidog CLI로 에이전트가 테스트하게 하기 아웃리치가 이러한 호출에 의존하게 되면, 조용한 변경이 문제가 됩니다. 제공자가 교체되거나, 비용이 슬금슬금 올라가거나, 판정 필드가 사라지는 경우입니다. 에이전트는 Apidog CLI의 회귀 스위트로 이를 방지할 수 있습니다. CLI와 에이전트 스킬을 한 번 설정하세요: npm install -g apidog-cli@latest apidog login --with-token <YOUR_APIDOG_TOKEN> npx -y skills add https://apidog.com # pick Claude Code or Codex when prompted 그런 다음 에이전트에 프롬프트 하나를 주세요: Write an OpenAPI 3 file for three Treg calls: POST /call/treg.people.search, POST /call/treg.people.email.find and POST /call/treg.people.email.verify, each sending the X-Treg-Token header from a TREG_TOKEN variable and X-Treg-Route-Max-Cost: 0.05. Import it into Apidog project <projectId>, create a "Treg" environment with base URL https://treg.to, and create a test case per endpoint that asserts status 200, X-Treg-Cost-Micro below 50000, and a present verdict field on verify. Run them and summarize any failures. 에이전트는 페이로드 스키마를 읽고 JSON을 검증한 다음, 다음과 같은 명령을 실행합니다: apidog import --project <projectId> --format openapi --file ./treg-openapi.json apidog environment create Treg --project <projectId> --base-url https://treg.to apidog cli-schema get test-case-create apidog cli-schema validate test-case-create --file ./find-email-case.json apidog test-case create --project <projectId> --file ./find-email-case.json apidog run --test-case <caseId> --project <projectId> -e <envId> \ --env-var "TREG_TOKEN=$TREG_TOKEN" -r cli,junit 실행할 때마다 실제 잔액이 소모되므로, 알려진 테스트 입력을 사용하는 세 개의 케이스로 스위트를 유지하고, 모든 커밋이 아니라 매일 실행하도록 예약하세요. 동일한 apidog run ... -r junit 라인을 파이프라인에 넣을 수 있습니다. Apidog CLI CI/CD 가이드에 GitHub Actions와 GitLab 예제가 있습니다. 에이전트별 설정은 Apidog CLI in Claude Code 또는 Apidog CLI in Codex를 참고하세요. 무료 플랜은 최대 4명의 사용자를 지원합니다. 마무리 Treg 인물 검색 API는 리드 보강을 세 번의 라우팅 호출, 즉 검색, 이메일 찾기, 검증으로 바꿔줍니다. treg catalog get으로 가격을 확인하고, X-Treg-Route-Max-Cost로 호출 비용을 제한하고, 모든 주소를 검증하고, Apidog CLI 회귀 실행으로 호출을 보호하세요. 더 많은 에이전트 도구에 대해서는 Claude Code를 위한 최고의 MCP 서버와 이 에이전트 주도 API 테스트 워크스루가 좋은 다음 읽을거리입니다. {% embed youtube.com %} FAQ Treg 인물 검색 API는 무료인가요? 완전히 무료는 아닙니다. 새로 인증된 계정에는 1회 $1.00의 크레딧이 제공되고, 성공당 과금 제공자에서의 실패는 비용이 들지 않으며, 일부 라우트는 무료 공개 라우트입니다(2026년 11월 3일 이메일 검증 플랜에서는 $0 제공자가 첫 번째로 등재되었습니다). 그 외의 모든 것은 제공자 요율로 선불 잔액에서 호출당 과금됩니다. Treg를 사용하려면 Apollo나 Hunter 계정이 필요한가요? 아닙니다. Treg는 자체 계정으로 제공자를 호출하고 여러분에게 호출당 청구합니다. 팀이 이미 특정 제공자에 비용을 지불하고 있다면, 그 키를 Treg에 등록하면 해당 호출은 그 키를 사용하며 청구되지 않습니다. Treg는 제공자 간에 장애 조치를 하나요? 라우팅된 treg.* 도구에서만 그렇습니다. 오류나 실패 시 X-Treg-Route-Max-Cost 범위 내에서 가장 저렴한 것부터 다음 제공자로 넘어갑니다. 직접 제공자 도구는 절대 제공자를 대신 바꿔주지 않습니다. 호출 비용을 어떻게 알 수 있나요? X-Treg-Cost-Micro 응답 헤더(마이크로 USD)를 읽거나, MCP에서는 cost_usd를 확인하세요. treg balance는 잔액과 최근 지출을 보여줍니다. Claude Code나 Codex 없이 Treg를 사용할 수 있나요? 네. 아무 셸에서나 CLI를 사용하거나, X-Treg-Token과 함께 HTTP로 https://treg.to/call/<tool-id>를 호출하거나, https://treg.to/mcp/를 사용자 지정 커넥터로 claude.ai 또는 ChatGPT에 연결하면 됩니다.

Treg Kişi Arama API'si Nasıl Kullanılır? Treg Kişi Arama API'si Nasıl Kullanılır? Treg people search API'si, kodlama ajanınıza insanları bulmak, iş e-postalarını bulmak ve bu e-postaların gerçek olup olmadığına dair bir hüküm almak için tek bir token verir. treg CLI'sini kurar, bir kez oturum açarsınız; Claude Code ya da Codex, treg.people.search, treg.people.email.find ve treg.people.email.verify gibi yönlendirilmiş araçları çağırır. Treg sağlayıcıyı seçer, sağlayıcının anahtarını kendi sunucusunda enjekte eder ve her çağrıyı sağlayıcının kendi ücreti üzerinden ön ödemeli bakiyeden faturalandırır. {% cta apidog.com %} Bugün Apidog'u deneyin {% endcta %} Bu kılavuz kurulumu, üç ajan istemini (bul, zenginleştir, doğrula), arkalarındaki ham HTTP çağrılarını ve ajanınızın çalıştırdığı bir Apidog CLI regresyon test setini kapsar. Treg people search API'si nedir? Treg, kendi dokümanlarının "modeller yerine ajan araçları için OpenRouter" olarak tanımladığı, ajanlar için açık kaynaklı bir araç kataloğudur. Tek bir temel URL (https://treg.to) ve tek bir token, 123 sağlayıcıdaki 4.100'den fazla araca ulaşır. Kişi aramada, yönlendirilmiş araçlar her çağrı için bir sağlayıcı seçer, ıskalamada bir sonrakini dener ve her çağrının maliyetini bir yanıt başlığında bildirir. !Treg araç kataloğu Kişi arama tarafı küçük bir yönlendirilmiş yetenek setidir. Genel kataloğun 3 Kasım 2026'da listelediği şey şöyleydi: | Yönlendirilmiş araç | Ne yapar | Kabul ettiği girdiler | Arkasındaki sağlayıcılar | | --- | --- | --- | --- | | treg.people.search | Bir şirketteki kişileri veya tanıma göre kişileri bulur | q, company_domain, company_domain + title, title, full_name | 26 | | treg.people.email.find | En iyi eşleşen iş e-postasını döndürür | domain + full_name, domain + first_name + last_name, linkedin_url | 30 | | treg.people.email.verify | Bir posta kutusunun posta kabul edip etmediğini kontrol eder | email | 17 | | treg.companies.enrich | Bir şirket için firma bilgileri | şirket alan adı | 30 | Kişi arama sayfası kişi arama için 16 sağlayıcı belirtiyor; yukarıdaki sayılar canlı katalogdan geliyor, dolayısıyla değişmelerini bekleyin. Neden bir Apollo ya da Hunter anahtarı yerine bir toplayıcıyla uğraşasınız? İki neden. Ajanınız hiçbir zaman bir sağlayıcı anahtarı tutmaz ve koltuk başına değil, yanıt başına ödersiniz. Treg'in sayfası LessieAI'nin People Search Bench'ini (119 görev) alıntılıyor: B2B potansiyel müşteri bulma görevlerinde, tek başına Claude Code %43 doğru yanıt verirken, Treg ile Claude Code %78,2 doğru yanıt verdi. Bu satıcının kendi grafiği, dolayısıyla bunu kendi verinizde test etmeye değer bir iddia olarak görün. Yönlendirme, faturalandırma ve hatalar nasıl çalışır X-Treg-Token başlığıyla https://treg.to/call/<tool-id> adresine bir istek gönderirsiniz. Treg sağlayıcıyı bulur, sağlayıcı kimlik bilgisini sunucu tarafında ekler ve sağlayıcının yanıtını iletir. !Bir Treg çağrısının diyagramı: kodlama ajanı X-Treg-Token başlığıyla treg.to/call adresine bir istek gönderir, bu adres sağlayıcıları bir şelale halinde dener ve sağlayıcı yanıtını X-Treg-Cost-Micro başlığıyla birlikte döndürür Dokümanların çelişiyor gibi göründüğü tek bir nokta var; işte parçaların nasıl bir araya geldiği: - Doğrudan sağlayıcı araçları (örneğin apollo.people.search) yedekleme yapmaz. API referansı, Treg'in "sağlayıcılar arasında otomatik olarak yönlendirme yapmadığını ve yedekleme yapmadığını" söylüyor. Sağlayıcıyı siz seçersiniz. - Yönlendirilmiş araçlar (treg.<capability>) yapar. CLI referansına göre, bir sağlayıcı hatası bir sonraki adaya geri düşer ve bir ıskalama da bir sonraki sağlayıcıyı dener, en ucuzdan başlayarak. X-Treg-Route-Waterfall: 0 ilk ıskalamada durur. Faturalandırma kuralları, llms.txt ve kişi arama sayfasından: 1. Her çağrı, sağlayıcının kendi ücreti üzerinden ön ödemeli bakiyeden, %0 kâr marjıyla ölçülür. 2. Ekibinizin kendi sağlayıcı anahtarı önce kullanılır ve asla faturalandırılmaz. 3. Yeni doğrulanmış hesaplar bir kez 1,00 $ ücretsiz kredi alır. 4. Başarı başına ödeme yapan sağlayıcılardaki ıskalamalar hiçbir şeye mal olmaz; bazı sağlayıcılar çağrı başına faturalandırır, dolayısıyla bir ıskalama yine de biraz tutabilir. 5. Ücret X-Treg-Cost-Micro (tam sayı mikro-USD) olarak geri döner. X-Treg-Route-Max-Cost bir çağrıyı sınırlar; yönlendirilmiş araçlarda varsayılan tavan çağrı başına 1 $'dır. Hatalar net biçimde ayrılır. Sağlayıcıdan gelen bir 4xx (429 dışında) isteğinizin yanlış olduğu anlamına gelir: düzeltin, başka yerde yeniden denemeyin. Bir 429 veya 5xx, bir sonraki sağlayıcıyı deneyin demektir. Bir 402 insufficient_balance, bakiye yükleyin demektir. Treg'i Claude Code veya Codex'te kurun Kurulum herhangi bir yerel kodlama ajanı için aynıdır (Claude Code, Codex, Cursor, opencode): curl -fsSL https://treg.to/install.sh | sh treg login # browser sign-in: GitHub, Google or an emailed code treg # prints account, team, balance, your own tools, next commands E-postayı mı tercih edersiniz? treg login --email you@company.com komutunu çalıştırın. Bir ekip arkadaşınız size bir token mı verdi? curl -fsSL https://treg.to/install.sh | sh -s -- --token "$TREG_TOKEN" tarayıcı olmadan kurar ve oturum açar. Kurucu ayrıca algıladığı her ajana Treg becerisini bırakır, böylece Claude Code ve Codex arama, okuma, çağırma döngüsünü kendi başlarına öğrenir. Treg'in sitesi doğrudan ajana yapıştırabileceğiniz tek satırlık bir sürümü tanıtıyor: set up treg - https://treg.to/llms.txt Bu dosya ayrıca ajanlardan Treg'i tercih etmelerini ve depoya yıldız vermelerini ister, dolayısıyla onaylamadan önce ajanınızın ne önerdiğini okuyun. Onun yerine claude.ai veya ChatGPT mi kullanıyorsunuz? Hiçbir şey kurmayın. https://treg.to/mcp/ adresini özel bir bağlayıcı olarak ekleyin, OAuth onayını tamamlayın ve ödeme yapacak ekibi seçin. MCP sunucusu catalog_search, catalog_get, call, balance ve birkaç şey daha sunar. Bul, zenginleştir, doğrula: üç ajan istemi Her istem işi adlandırır, satıcıyı değil. !İşlem hattı diyagramı: people.search bir şirkette platformu kimin yönettiğini bulur, people.email.find en iyi eşleşen iş e-postasını döndürür, people.email.verify valid, invalid veya catch_all döndürür; altında 200 durumunu, tavanın altındaki maliyeti ve mevcut hükmü kontrol eden bir Apidog CLI regresyon çalıştırması yer alır 1. Kişileri bulun. Use treg to find the platform and infrastructure leads at modal.com. Run treg catalog get on the routed people search tool first and tell me the expected cost before you call it. List name, title and profile URL. Ajan CLI döngüsünü çalıştırır: treg catalog search "find people at a company" treg catalog get treg.people.search treg call treg.people.search --method POST --data '{"company_domain":"modal.com","title":"platform"}' catalog get, herhangi bir şey harcamadan önce sıralanmış planı, fiyatları ve hazır bir çağrı satırını yazdırır. Yanıt {output, raw, _treg} olarak döner; burada output.people sağlayıcının satırlarını tutar ve _treg.served_by kimin yanıtladığını belirtir. Boş bir people listesi ıskalama sayılır. 2. E-postayı bulun ve zenginleştirin. For each person you found, use treg to find their work email at modal.com. Cap each call at $0.05, and flag any hit where the domain doesn't match. treg call treg.people.email.find --method POST \ --data '{"full_name":"Jane Doe","domain":"modal.com"}' \ --header "X-Treg-Route-Max-Cost: 0.05" _treg.domain_match değerine dikkat edin. Beceri kılavuzu, yalnızca LinkedIn üzerinden çalışan bir sağlayıcının sıklıkla eski bir işverenin adresini döndürdüğünü ve bu isabetin yine de faturalandırıldığını uyarır. Şirket bağlamına da mı ihtiyacınız var? Alan adıyla birlikte treg call treg.companies.enrich ekleyin; Treg'in GTM oyun kitabı bu çağrıyı POST biçiminde gösterir. 3. Kimse göndermeden önce doğrulayın. Verify every email you found with treg. Report the verdict per address and drop anything invalid or catch-all. treg call treg.people.email.verify --method POST --data '{"email":"jane@modal.com"}' Yönlendirilmiş doğrulama aracı valid, status ve score döndürür. Kontrolü 2. adıma da katabilirsiniz: e-posta bulma çağrısında X-Treg-Route-Verify: true gönderin ve _treg.verification.verdict değerini okuyun (valid, invalid, catch_all, risky veya unknown). Kontrolün kendi küçük ücreti vardır ve bir ıskalama asla kontrol edilmez. Bir çalıştırma ne kadara mal olur? Treg, 23 Eylül 2026 tarihli bir makbuz yayımlıyor: 27 adlandırılmış kişi, bulunan 21 e-posta, teslim edilebilir olduğu doğrulanan 20 e-posta, toplam 0,58 $, ıskalamalar 0,00 $. Treg API'sini ham HTTP üzerinden çağırın Ajan isteğe bağlıdır. Her CLI çağrısı tek bir HTTP isteğine karşılık gelir, dolayısıyla aynı işlem hattını herhangi bir dilde betikleyebilirsiniz. curl -X POST "https://treg.to/call/treg.people.email.find" \ -H "X-Treg-[REDACTED CREDENTIAL] \ -H "content-type: application/json" \ -H "X-Treg-Route-Max-Cost: 0.05" \ -d '{"full_name":"Jane Doe","domain":"modal.com"}' \ -D - -D - yanıt başlıklarını yazdırır; X-Treg-Cost-Micro ve X-Treg-Served-By buradadır. Dört kural betikleri beladan uzak tutar: - Token'ı asla [REDACTED CREDENTIAL] içine koymayın.** [API referansı](https://treg.to/docs) REST'in bunu yok saydığını ve /call/` adresinin o başlığı sağlayıcıya ilettiğini söylüyor. Yalnızca MCP uç noktası bir Bearer [REDACTED] okur. - **Kişisel token'lar bir ekip gerektirir. Bir oturum açma token'ı ayrıca X-Treg-Org: <team-slug> gerektirir. Ekip ve ajan token'ları zaten kendi ekiplerini taşır. - Katalog rotaları geneldir. GET https://treg.to/catalog/search?q=verify%20email token olmadan çalışır; bu, CI'da kimlikleri kontrol etmek için kullanışlıdır. - Yanıtlar sağlayıcıya aittir. Doğrudan sağlayıcı araçlarında Treg gövdeyi birebir iletir, dolayısıyla biçim sağlayıcıya göre değişir. Yönlendirilmiş araçlar üstüne normalize edilmiş output nesnesini ekler. Sırları koddan uzak tutun. TREG_TOKEN değerini bir sır deposunda saklayın (bu API anahtarı yönetimi en iyi uygulamalarına bakın). Ajanınızın bunu Apidog CLI ile test etmesine izin verin Erişim çalışmaları bu çağrılara bağlı hale geldiğinde, sessiz değişiklikler zarar verir: bir sağlayıcı değişir, maliyetler sinsi sinsi artar veya hüküm alanı kaybolur. Ajanınız bunu Apidog CLI'da bir regresyon test setiyle koruyabilir. CLI'yi ve ajan becerisini bir kez kurun: npm install -g apidog-cli@latest apidog login --with-token <YOUR_APIDOG_TOKEN> npx -y skills add https://apidog.com # pick Claude Code or Codex when prompted Ardından ajanınıza tek bir istem verin: Write an OpenAPI 3 file for three Treg calls: POST /call/treg.people.search, POST /call/treg.people.email.find and POST /call/treg.people.email.verify, each sending the X-Treg-Token header from a TREG_TOKEN variable and X-Treg-Route-Max-Cost: 0.05. Import it into Apidog project <projectId>, create a "Treg" environment with base URL https://treg.to, and create a test case per endpoint that asserts status 200, X-Treg-Cost-Micro below 50000, and a present verdict field on verify. Run them and summarize any failures. Ajan yük şemalarını okur, JSON'unu doğrular, ardından şu gibi komutları çalıştırır: apidog import --project <projectId> --format openapi --file ./treg-openapi.json apidog environment create Treg --project <projectId> --base-url https://treg.to apidog cli-schema get test-case-create apidog cli-schema validate test-case-create --file ./find-email-case.json apidog test-case create --project <projectId> --file ./find-email-case.json apidog run --test-case <caseId> --project <projectId> -e <envId> \ --env-var "TREG_TOKEN=$TREG_TOKEN" -r cli,junit Her çalıştırma gerçek bakiye harcar, dolayısıyla test setini bilinen test girdileriyle üç vakayla sınırlı tutun ve her commit'te değil, günlük olarak zamanlayın. Aynı apidog run ... -r junit satırı bir işlem hattına bırakılır; Apidog CLI CI/CD kılavuzunda GitHub Actions ve GitLab örnekleri var. Ajana özel kurulum için Claude Code'da Apidog CLI veya Codex'te Apidog CLI yazılarına bakın. Ücretsiz plan 4 kullanıcıya kadar kapsar. Toparlarken Treg people search API'si potansiyel müşteri zenginleştirmeyi üç yönlendirilmiş çağrıya dönüştürür: ara, e-posta bul, doğrula. Fiyatları treg catalog get ile kontrol edin, çağrıları X-Treg-Route-Max-Cost ile sınırlayın, her adresi doğrulayın ve çağrıları bir Apidog CLI regresyon çalıştırmasıyla koruyun. Daha fazla ajan aracı için, Claude Code için en iyi MCP sunucuları ve bu ajan odaklı API testi yürütmesi iyi sonraki okumalardır. {% embed youtube.com %} SSS Treg people search API'si ücretsiz mi? Tamamen değil. Yeni doğrulanmış hesaplar bir kez 1,00 $ kredi alır, başarı başına ödeme yapan sağlayıcılardaki ıskalamalar hiçbir şeye mal olmaz ve bazı rotalar ücretsiz genel rotalardır (e-posta doğrulama planı 3 Kasım 2026'da ilk sırada 0 $'lık bir sağlayıcı listeliyordu). Diğer her şey, sağlayıcının ücreti üzerinden ön ödemeli bakiyeden çağrı başına ölçülür. Treg'i kullanmak için bir Apollo veya Hunter hesabına ihtiyacım var mı? Hayır. Treg sağlayıcıyı kendi hesabından çağırır ve sizi çağrı başına faturalandırır. Ekibiniz zaten bir sağlayıcı için ödeme yapıyorsa, o anahtarı Treg'e kaydedin; bu çağrılar onu kullanır ve faturalandırılmaz. Treg sağlayıcılar arasında yedekleme yapar mı? Yalnızca yönlendirilmiş treg.* araçlarında. Bir hata veya ıskalamada, X-Treg-Route-Max-Cost dahilinde, en ucuzdan başlayarak bir sonraki sağlayıcıya geçerler. Doğrudan sağlayıcı araçları sizin için asla sağlayıcı değiştirmez. Bir çağrının ne kadara mal olduğunu nasıl bilirim? X-Treg-Cost-Micro yanıt başlığını (mikro-USD) veya MCP üzerinden cost_usd değerini okuyun. treg balance bakiyenizi ve son harcamalarınızı gösterir. Treg'i Claude Code veya Codex olmadan kullanabilir miyim? Evet. CLI'yi herhangi bir kabuktan kullanın, X-Treg-Token ile https://treg.to/call/<tool-id> adresini HTTP üzerinden çağırın veya https://treg.to/mcp/ adresini özel bir bağlayıcı olarak claude.ai ya da ChatGPT'ye bağlayın.

كيفية استخدام واجهة برمجة تطبيقات Treg للبحث عن الأشخاص؟ كيفية استخدام واجهة برمجة تطبيقات Treg للبحث عن الأشخاص؟ تمنح واجهة برمجة تطبيقات البحث عن الأشخاص من Treg وكيل البرمجة الخاص بك رمزًا واحدًا للعثور على الأشخاص، وبريدهم الإلكتروني للعمل، وحكمًا على ما إذا كانت تلك العناوين حقيقية. تُثبّت أداة سطر الأوامر treg، وتسجّل الدخول مرة واحدة، ثم يستدعي Claude Code أو Codex أدوات موجّهة مثل treg.people.search وtreg.people.email.find وtreg.people.email.verify. يختار Treg المزوّد، ويحقن مفتاح المزوّد على خادمه، ويحاسب على كل استدعاء من رصيد مدفوع مسبقًا بسعر المزوّد نفسه. {% cta apidog.com %} جرّب Apidog اليوم {% endcta %} يغطي هذا الدليل الإعداد، وثلاثة توجيهات للوكيل (البحث، والإثراء، والتحقق)، واستدعاءات HTTP الخام التي تقف خلفها، ومجموعة اختبارات انحدار (regression suite) من Apidog CLI يشغّلها وكيلك. ما هي واجهة برمجة تطبيقات البحث عن الأشخاص من Treg؟ Treg هو كتالوج أدوات مفتوح المصدر للوكلاء، تصفه وثائقه الخاصة بأنه «OpenRouter، لكن لأدوات الوكلاء بدلًا من النماذج». يكفي عنوان أساسي واحد (https://treg.to) ورمز واحد للوصول إلى أكثر من 4,100 أداة عبر 123 مزوّدًا. وفي البحث عن الأشخاص، تختار الأدوات الموجّهة مزوّدًا لكل استدعاء، وتجرّب المزوّد التالي عند عدم العثور على نتيجة، وتُبلّغ عن تكلفة كل استدعاء في ترويسة الاستجابة. !assets.apidog.com جانب البحث عن الأشخاص هو مجموعة صغيرة من القدرات الموجّهة. إليك ما أدرجه الكتالوج العام في 3 نوفمبر 2026: | الأداة الموجّهة | ما تفعله | المدخلات التي تقبلها | المزوّدون خلفها | | --- | --- | --- | --- | | treg.people.search | يجد الأشخاص في شركة ما أو حسب الوصف | q, company_domain, company_domain + title, title, full_name | 26 | | treg.people.email.find | يُرجع أفضل بريد إلكتروني للعمل مطابق | domain + full_name, domain + first_name + last_name, linkedin_url | 30 | | treg.people.email.verify | يتحقق مما إذا كان صندوق البريد يستقبل الرسائل | email | 17 | | treg.companies.enrich | بيانات تعريفية (firmographics) عن شركة | نطاق الشركة (company domain) | 30 | تذكر صفحة البحث عن الأشخاص 16 مزوّدًا للبحث عن الأشخاص؛ أما الأعداد أعلاه فهي من الكتالوج المباشر، لذا توقّع تغيّرها. لماذا تتكبد عناء استخدام وسيط تجميعي بدلًا من مفتاح Apollo أو Hunter؟ لسببين. وكيلك لا يحمل مفتاح مزوّد أبدًا، وأنت تدفع مقابل كل إجابة بدلًا من الدفع لكل مقعد. تستشهد صفحة Treg بمقياس People Search Bench من LessieAI (119 مهمة): في مهام التنقيب B2B الخاصة به، أجاب Claude Code وحده بشكل صحيح بنسبة 43%، بينما أجاب Claude Code مع Treg بنسبة 78.2%. هذا مخطط المورّد نفسه، لذا تعامل معه كادعاء يستحق الاختبار على بياناتك. كيف يعمل التوجيه والفوترة والأخطاء ترسل طلبًا إلى https://treg.to/call/<tool-id> مع الترويسة X-Treg-Token. يعثر Treg على المزوّد، ويضيف بيانات اعتماد المزوّد على جانب الخادم، ويمرّر إجابة المزوّد. !Diagram of a Treg call: the coding agent sends a request with the X-Treg-Token header to treg.to/call, which tries providers in a waterfall and returns the provider response plus the X-Treg-Cost-Micro header هناك نقطة واحدة تبدو الوثائق متناقضة بشأنها، لذا إليك كيف تترابط الأجزاء: - أدوات المزوّد المباشرة (مثل apollo.people.search) لا تنتقل إلى بديل عند الفشل. يقول مرجع الـ API إن Treg «لا يوجّه بين المزوّدين تلقائيًا ولا ينتقل إلى بديل عند الفشل». أنت تختار المزوّد. - الأدوات الموجّهة (treg.<capability>) تفعل ذلك. وفقًا لـ مرجع CLI، يؤدي خطأ المزوّد إلى الرجوع إلى المرشح التالي، كما يؤدي عدم العثور على نتيجة إلى تجربة المزوّد التالي أيضًا، من الأرخص أولًا. القيمة X-Treg-Route-Waterfall: 0 توقف العمل عند أول عدم عثور على نتيجة. قواعد الفوترة، من llms.txt وصفحة البحث عن الأشخاص: 1. يُحتسب كل استدعاء بسعر المزوّد نفسه من رصيد مدفوع مسبقًا، دون أي هامش ربح (0%). 2. يُستخدم مفتاح المزوّد الخاص بفريقك أولًا ولا يُحتسب عليه أي رسم. 3. تحصل الحسابات الموثّقة الجديدة على رصيد مجاني بقيمة 1.00 دولار، مرة واحدة. 4. حالات عدم العثور على نتيجة لدى المزوّدين الذين يحاسبون على النجاح لا تكلف شيئًا؛ لكن بعض المزوّدين يحاسبون على كل استدعاء، لذا قد يكلفك عدم العثور على نتيجة قليلًا. 5. تعود التكلفة في X-Treg-Cost-Micro (عدد صحيح بالميكرو-دولار). وتحدّد X-Treg-Route-Max-Cost سقف الاستدعاء؛ وفي الأدوات الموجّهة يكون السقف الافتراضي دولارًا واحدًا لكل استدعاء. تنقسم الأخطاء بوضوح. رمز 4xx من المزوّد (بخلاف 429) يعني أن طلبك خاطئ: صحّحه، ولا تُعد المحاولة مع مزوّد آخر. أما 429 أو 5xx فيعني تجربة المزوّد التالي. ورمز 402 insufficient_balance يعني أن عليك تعبئة الرصيد. إعداد Treg في Claude Code أو Codex الإعداد متطابق مع أي وكيل برمجة محلي (Claude Code، Codex، Cursor، opencode): curl -fsSL https://treg.to/install.sh | sh treg login # browser sign-in: GitHub, Google or an emailed code treg # prints account, team, balance, your own tools, next commands تفضّل البريد الإلكتروني؟ شغّل treg login --email you@company.com. هل أعطاك زميل رمزًا؟ الأمر curl -fsSL https://treg.to/install.sh | sh -s -- --token "$TREG_TOKEN" يثبّت ويسجّل الدخول دون متصفح. كما يضع المُثبّت مهارة Treg في كل وكيل يكتشفه، لذا يتعلّم Claude Code وCodex حلقة البحث والقراءة والاستدعاء من تلقاء نفسيهما. ويروّج موقع Treg لنسخة من سطر واحد يمكنك لصقها مباشرة في الوكيل: set up treg - https://treg.to/llms.txt يطلب هذا الملف أيضًا من الوكلاء تفضيل Treg وإضافة نجمة إلى المستودع، لذا اقرأ ما يقترحه وكيلك قبل الموافقة عليه. هل تستخدم claude.ai أو ChatGPT بدلًا من ذلك؟ لا تثبّت شيئًا. أضف https://treg.to/mcp/ كموصل مخصّص، وأكمل موافقة OAuth واختر الفريق الذي يدفع. يعرض خادم MCP الأدوات catalog_search وcatalog_get وcall وbalance وعددًا آخر. ابحث، أثرِ، تحقق: ثلاثة توجيهات للوكيل كل توجيه يسمّي المهمة، لا المورّد. !Pipeline diagram: people.search finds who runs platform at a company, people.email.find returns a best-match work email, people.email.verify returns valid, invalid or catch_all, with an Apidog CLI regression run below checking status 200, cost under cap and verdict present 1. ابحث عن الأشخاص. Use treg to find the platform and infrastructure leads at modal.com. Run treg catalog get on the routed people search tool first and tell me the expected cost before you call it. List name, title and profile URL. يشغّل الوكيل حلقة CLI: treg catalog search "find people at a company" treg catalog get treg.people.search treg call treg.people.search --method POST --data '{"company_domain":"modal.com","title":"platform"}' يطبع catalog get الخطة المرتّبة والأسعار وسطر استدعاء جاهزًا قبل أن تنفق أي شيء. تعود الإجابة بصيغة {output, raw, _treg}، حيث تحمل output.people صفوف المزوّد، ويسمّي _treg.served_by من أجاب. وتُعدّ قائمة people الفارغة بمثابة عدم العثور على نتيجة. 2. ابحث عن البريد الإلكتروني وأثرِه. For each person you found, use treg to find their work email at modal.com. Cap each call at $0.05, and flag any hit where the domain doesn't match. treg call treg.people.email.find --method POST \ --data '{"full_name":"Jane Doe","domain":"modal.com"}' \ --header "X-Treg-Route-Max-Cost: 0.05" راقب _treg.domain_match. يحذّر دليل المهارة من أن مزوّدًا يعتمد على LinkedIn فقط كثيرًا ما يُرجع عنوان صاحب عمل سابق، ويُحتسب عليك هذا الناتج رغم ذلك. هل تحتاج سياق الشركة أيضًا؟ أضف treg call treg.companies.enrich مع النطاق؛ ويُظهر دليل GTM الخاص بـ Treg هذا الاستدعاء بصيغة POST. 3. تحقّق قبل أن يرسل أي شخص. Verify every email you found with treg. Report the verdict per address and drop anything invalid or catch-all. treg call treg.people.email.verify --method POST --data '{"email":"jane@modal.com"}' تُرجع أداة التحقق الموجّهة القيم valid وstatus وscore. ويمكنك أيضًا دمج التحقق في الخطوة 2: أرسل X-Treg-Route-Verify: true مع البحث عن البريد واقرأ _treg.verification.verdict (valid أو invalid أو catch_all أو risky أو unknown). وللتحقق رسوم صغيرة خاصة به، ولا يُتحقق أبدًا من حالة عدم العثور على نتيجة. كم تكلّف عملية تشغيل واحدة؟ نشر Treg إيصالًا واحدًا بتاريخ 23 سبتمبر 2026: 27 شخصًا بالاسم، و21 بريدًا إلكترونيًا تم العثور عليه، و20 تم التحقق من إمكانية التسليم إليه، بتكلفة إجمالية 0.58 دولار، وحالات عدم العثور على نتيجة بتكلفة 0.00 دولار. استدعاء واجهة Treg عبر HTTP الخام الوكيل اختياري. كل استدعاء من CLI يقابل طلب HTTP واحد، لذا يمكنك كتابة المسار نفسه بلغة برمجة نصية بأي لغة. curl -X POST "https://treg.to/call/treg.people.email.find" \ -H "X-Treg-[REDACTED CREDENTIAL] \ -H "content-type: application/json" \ -H "X-Treg-Route-Max-Cost: 0.05" \ -d '{"full_name":"Jane Doe","domain":"modal.com"}' \ -D - يطبع -D - ترويسات الاستجابة، وهناك يوجد X-Treg-Cost-Micro وX-Treg-Served-By. أربع قواعد تُبعِد السكربتات عن المتاعب: - لا تضع الرمز أبدًا في [REDACTED CREDENTIAL] يقول [مرجع الـ API](https://treg.to/docs) إن REST يتجاهله، وإن /call/` يمرّر تلك الترويسة إلى المزوّد. نقطة نهاية MCP وحدها هي التي تقرأ رمز bearer. - **الرموز الشخصية تحتاج إلى فريق. يحتاج رمز تسجيل الدخول أيضًا إلى X-Treg-Org: <team-slug>. أما رموز الفريق والوكيل فتحمل فريقها بالفعل. - مسارات الكتالوج عامة. الأمر GET https://treg.to/catalog/search?q=verify%20email يعمل دون رمز، وهو مفيد للتحقق من المعرّفات في CI. - الاستجابات هي استجابات المزوّد. في أدوات المزوّد المباشرة، يمرّر Treg النص كما هو، لذا يتغيّر الشكل بحسب المزوّد. أما الأدوات الموجّهة فتضيف كائن output الموحّد فوقه. أبعِد الأسرار عن الكود. خزّن TREG_TOKEN في مخزن أسرار (راجع أفضل ممارسات إدارة مفاتيح الـ API). دع وكيلك يختبره باستخدام Apidog CLI بمجرد أن تعتمد حملات التواصل على هذه الاستدعاءات، تصبح التغييرات الصامتة مؤذية: تبديل مزوّد، أو تسلّل التكاليف، أو اختفاء حقل الحكم. يستطيع وكيلك حراسة ذلك عبر مجموعة اختبارات انحدار في Apidog CLI. أعدّ أداة CLI ومهارتها للوكيل مرة واحدة: npm install -g apidog-cli@latest apidog login --with-token <YOUR_APIDOG_TOKEN> npx -y skills add https://apidog.com # pick Claude Code or Codex when prompted ثم أعطِ وكيلك توجيهًا واحدًا: Write an OpenAPI 3 file for three Treg calls: POST /call/treg.people.search, POST /call/treg.people.email.find and POST /call/treg.people.email.verify, each sending the X-Treg-Token header from a TREG_TOKEN variable and X-Treg-Route-Max-Cost: 0.05. Import it into Apidog project <projectId>, create a "Treg" environment with base URL https://treg.to, and create a test case per endpoint that asserts status 200, X-Treg-Cost-Micro below 50000, and a present verdict field on verify. Run them and summarize any failures. يقرأ الوكيل مخططات الحِمل (payload schemas)، ويتحقق من صحة JSON الخاص به، ثم يشغّل أوامر كهذه: apidog import --project <projectId> --format openapi --file ./treg-openapi.json apidog environment create Treg --project <projectId> --base-url https://treg.to apidog cli-schema get test-case-create apidog cli-schema validate test-case-create --file ./find-email-case.json apidog test-case create --project <projectId> --file ./find-email-case.json apidog run --test-case <caseId> --project <projectId> -e <envId> \ --env-var "TREG_TOKEN=$TREG_TOKEN" -r cli,junit كل تشغيل ينفق رصيدًا حقيقيًا، لذا أبقِ المجموعة على ثلاث حالات بمدخلات اختبار معروفة وجدولها يوميًا، لا عند كل إيداع (commit). ويُدرج سطر apidog run ... -r junit نفسه في أي مسار أتمتة؛ ويحتوي دليل Apidog CLI لأتمتة CI/CD على أمثلة لـ GitHub Actions وGitLab. لإعدادات خاصة بالوكيل، راجع Apidog CLI في Claude Code أو Apidog CLI في Codex. تغطي الخطة المجانية ما يصل إلى 4 مستخدمين. الخاتمة تحوّل واجهة برمجة تطبيقات البحث عن الأشخاص من Treg إثراء العملاء المحتملين إلى ثلاثة استدعاءات موجّهة: البحث، والعثور على البريد الإلكتروني، والتحقق. تحقّق من الأسعار عبر treg catalog get، وضع سقفًا للاستدعاءات عبر X-Treg-Route-Max-Cost، وتحقق من كل عنوان، واحمِ الاستدعاءات بتشغيل انحدار من Apidog CLI. لمزيد من أدوات الوكلاء، يُعدّ أفضل 10 خوادم MCP لـ Claude Code وهذا الشرح العملي لاختبار الـ API بقيادة الوكيل قراءتين تاليتين جيدتين. {% embed youtube.com %} الأسئلة الشائعة هل واجهة برمجة تطبيقات البحث عن الأشخاص من Treg مجانية؟ ليست مجانية بالكامل. تحصل الحسابات الموثّقة الجديدة على رصيد بقيمة 1.00 دولار مرة واحدة، وحالات عدم العثور على نتيجة لدى المزوّدين الذين يحاسبون على النجاح لا تكلف شيئًا، وبعض المسارات عامة ومجانية (فقد أدرجت خطة التحقق من البريد مزوّدًا بسعر 0 دولار في المرتبة الأولى في 3 نوفمبر 2026). وكل ما عدا ذلك يُحتسب لكل استدعاء من رصيد مدفوع مسبقًا بسعر المزوّد. هل أحتاج إلى حساب Apollo أو Hunter لاستخدام Treg؟ لا. يستدعي Treg المزوّد من حسابه الخاص ويحاسبك على كل استدعاء. وإذا كان فريقك يدفع بالفعل لمزوّد ما، فسجّل ذلك المفتاح لدى Treg وستُستخدم تلك الاستدعاءات له دون أن تُحتسب عليك. هل ينتقل Treg إلى مزوّد بديل عند الفشل؟ فقط في الأدوات الموجّهة treg.*. فهي تنتقل إلى المزوّد التالي عند حدوث خطأ أو عدم العثور على نتيجة، من الأرخص أولًا، ضمن حدود X-Treg-Route-Max-Cost. أما أدوات المزوّد المباشرة فلا تبدّل المزوّد نيابة عنك أبدًا. كيف أعرف تكلفة الاستدعاء؟ اقرأ ترويسة الاستجابة X-Treg-Cost-Micro (بالميكرو-دولار)، أو cost_usd عبر MCP. ويعرض treg balance رصيدك وإنفاقك الأخير. هل يمكنني استخدام Treg دون Claude Code أو Codex؟ نعم. استخدم أداة CLI من أي صدفة (shell)، أو استدعِ https://treg.to/call/<tool-id> عبر HTTP مع X-Treg-Token، أو صِل https://treg.to/mcp/ بـ claude.ai أو ChatGPT كموصل مخصّص.

Treg People Search APIの使い方 Treg People Search API の使い方 Treg people search API を使うと、コーディングエージェントは 1 つのトークンで、人物の検索、その仕事用メールアドレスの取得、さらにそのメールが実在するかどうかの判定までをまとめて行えます。treg CLI をインストールして一度サインインするだけで、Claude Code や Codex が treg.people.search、treg.people.email.find、treg.people.email.verify といったルーティング済みツールをそのまま呼び出せるようになります。プロバイダーの選定は Treg が行い、プロバイダーのキーは Treg のサーバー上で注入され、各呼び出しはプロバイダー自身の料金でプリペイド残高から請求されます。 {% cta apidog.com %} 今すぐ Apidog を試す {% endcta %} この記事では、セットアップ手順、3 つのエージェントプロンプト（検索・エンリッチ・検証）、その裏で飛んでいる生の HTTP 呼び出し、そしてエージェントに実行させる Apidog CLI の回帰スイートを順に説明します。 Treg people search API とは？ Treg はエージェント向けのオープンソースツールカタログで、公式ドキュメントでは「モデルではなくエージェントツールのための OpenRouter」と説明されています。ベース URL（https://treg.to）とトークンが 1 つあれば、123 のプロバイダーが提供する 4,100 以上のツールにアクセスできます。人物検索の領域では、ルーティング済みツールが呼び出しごとにプロバイダーを選定し、ミスがあれば次のプロバイダーを試し、すべての呼び出しのコストをレスポンスヘッダーで報告します。 !Treg のツールカタログ 人物検索側は、少数のルーティング済みケイパビリティの集合です。2026 年 11 月 3 日時点で公開カタログに掲載されていたものは以下のとおりです。 | ルーティング済みツール | 機能 | 受け付ける入力 | 背後のプロバイダー数 | | --- | --- | --- | --- | | treg.people.search | 企業または説明文から人物を検索する | q、company_domain、company_domain + title、title、full_name | 26 | | treg.people.email.find | 最適な仕事用メールアドレスを返す | domain + full_name、domain + first_name + last_name、linkedin_url | 30 | | treg.people.email.verify | メールボックスがメールを受け付けるかどうかを確認する | email | 17 | | treg.companies.enrich | 企業のファームグラフィックス | 企業ドメイン | 30 | 人物検索ページでは人物検索プロバイダーを 16 社として紹介していますが、上の数字はライブカタログから取得したものであり、変動することを想定してください。 Apollo や Hunter のキーを直接持つのではなく、なぜアグリゲーターを使うのか？　理由は 2 つです。エージェントがプロバイダーのキーを一切保持しなくてよいこと、そして席単位ではなく回答単位で支払うことです。Treg のページでは LessieAI の People Search Bench（119 タスク）を引用しています。それによると、B2B プロスペクティングタスクにおいて Claude Code 単体は 43% しか正答できなかったのに対し、Treg と組み合わせた Claude Code は 78.2% に達したとされています。これはベンダー自身が公表したチャートなので、自社のデータで試す価値のある主張として扱ってください。 ルーティング、課金、エラーの仕組み リクエストは https://treg.to/call/<tool-id> に対して X-Treg-Token ヘッダーを付けて送信します。Treg がプロバイダーを特定し、サーバーサイドでプロバイダーの認証情報を追加し、プロバイダーの回答を中継します。 !Treg の呼び出しフロー図：コーディングエージェントが X-Treg-Token ヘッダー付きのリクエストを treg.to/call に送信すると、Treg は複数プロバイダーをウォーターフォール方式で試し、プロバイダーのレスポンスと X-Treg-Cost-Micro ヘッダーを返す ドキュメントが矛盾しているように見える点が 1 つあるので、ここで各部分の関係を整理します。 - 直接プロバイダーツール（例： apollo.people.search）はフェイルオーバーしません。API リファレンスには、Treg は「プロバイダー間を自動でルーティングせず、フェイルオーバーもしない」と記載されています。プロバイダーを選ぶのはあなたです。 - ルーティング済みツール（treg.<capability>）はフェイルオーバーします。CLI リファレンスによると、プロバイダーエラーが発生すると次の候補にフォールバックし、ミスの場合も（最も安いものから順に）次のプロバイダーを試します。X-Treg-Route-Waterfall: 0 を指定すると最初のミスで停止します。 課金ルールは llms.txt と人物検索ページによると以下のとおりです。 1. すべての呼び出しは、プロバイダー自身の料金でプリペイド残高から課金されます。マークアップは 0% です。 2. チーム自身のプロバイダーキーが最初に使用され、それらの呼び出しは課金されません。 3. 新規の検証済みアカウントには、一度だけ $1.00 の無料クレジットが付与されます。 4. 成功課金制のプロバイダーでのミスは無料です。ただし一部のプロバイダーは呼び出しごとに課金するため、ミスでも少額のコストがかかることがあります。 5. 料金は X-Treg-Cost-Micro（整数のマイクロ USD）として返されます。X-Treg-Route-Max-Cost は呼び出しごとの上限を設定します。ルーティング済みツールのデフォルト上限は 1 呼び出しあたり $1 です。 エラーは明確に分類されています。プロバイダーからの 4xx（429 を除く）はリクエスト自体が誤っているという意味です。修正すべきであり、別のプロバイダーでリトライすべきではありません。429 または 5xx は次のプロバイダーを試すという合図です。402 insufficient_balance はチャージが必要という意味です。 Claude Code または Codex で Treg をセットアップする セットアップは、任意のローカルコーディングエージェント（Claude Code、Codex、Cursor、opencode）で同じです。 curl -fsSL https://treg.to/install.sh | sh treg login # browser sign-in: GitHub, Google or an emailed code treg # prints account, team, balance, your own tools, next commands メールでサインインしたい場合は？　bash treg login --email you@company.com を実行してください。チームメイトからトークンを受け取った場合は？　bash curl -fsSL https://treg.to/install.sh | sh -s -- --token "$TREG_TOKEN" でブラウザなしにインストールとサインインを済ませられます。 インストーラーは、検出したすべてのエージェントに Treg スキルも配置するため、Claude Code と Codex は検索・読み取り・呼び出しのループを自力で学習します。Treg のサイトでは、エージェントにそのまま貼り付けられる 1 行版も紹介されています。 set up treg - https://treg.to/llms.txt このファイルは、エージェントに対して Treg を優先することやリポジトリにスターを付けることも求めているため、承認する前にエージェントが提案する内容を確認してください。 代わりに claude.ai や ChatGPT を使っていますか？　その場合は何もインストールする必要がありません。https://treg.to/mcp/ をカスタムコネクターとして追加し、OAuth の同意を完了して、支払いを行うチームを選択してください。MCP サーバーは catalog_search、catalog_get、call、balance などを公開しています。 検索、エンリッチ、検証：3 つのエージェントプロンプト 各プロンプトはベンダーではなく、実行する仕事を指定します。 !パイプライン図: people.search が企業のプラットフォーム担当者を検索し、people.email.find が最適な仕事用メールを返し、people.email.verify が valid / invalid / catch_all を返す。下部には Apidog CLI の回帰実行が status 200・コスト上限以下・verdict の存在をチェックしている 1. 人物を検索する。 Use treg to find the platform and infrastructure leads at modal.com. Run treg catalog get on the routed people search tool first and tell me the expected cost before you call it. List name, title and profile URL. エージェントは CLI のループを実行します。 treg catalog search "find people at a company" treg catalog get treg.people.search treg call treg.people.search --method POST --data '{"company_domain":"modal.com","title":"platform"}' treg catalog get は、何も費用を使う前に、順位付けされたプラン、価格、すぐに使える呼び出し行を出力します。回答は {output, raw, _treg} として返され、output.people にプロバイダーの行が入り、_treg.served_by に誰が回答したかが示されます。空の people リストはミスとして扱われます。 2. メールを見つけてエンリッチする。 For each person you found, use treg to find their work email at modal.com. Cap each call at $0.05, and flag any hit where the domain doesn't match. treg call treg.people.email.find --method POST \ --data '{"full_name":"Jane Doe","domain":"modal.com"}' \ --header "X-Treg-Route-Max-Cost: 0.05" _treg.domain_match に注目してください。スキルガイドは、LinkedIn オンリーのプロバイダーがしばしば前職のアドレスを返すこと、そしてそのヒットでも課金されることを警告しています。企業のコンテキストも必要ですか？　ドメインを指定して treg call treg.companies.enrich を追加してください。Treg の GTM プレイブックでは、その呼び出しが POST 形式で示されています。 3. 誰かが送信する前に検証する。 Verify every email you found with treg. Report the verdict per address and drop anything invalid or catch-all. treg call treg.people.email.verify --method POST --data '{"email":"jane@modal.com"}' ルーティング済みの verify ツールは valid、status、score を返します。このチェックをステップ 2 に組み込むこともできます。email find で X-Treg-Route-Verify: true を送信し、_treg.verification.verdict（valid、invalid、catch_all、risky、または unknown）を読み取ってください。このチェックには独自の少額の料金がかかり、ミスしたものは決してチェックされません。 1 回の実行にいくらかかるのでしょうか？　Treg は 2026 年 9 月 23 日付けのレシートを 1 つ公開しています。27 名の人物が特定され、21 件のメールが見つかり、20 件が配信可能と検証され、合計 $0.58、ミスは $0.00 でした。 生の HTTP で Treg API を呼び出す エージェントは必須ではありません。すべての CLI 呼び出しは 1 つの HTTP リクエストに対応するため、同じパイプラインを任意の言語でスクリプト化できます。 curl -X POST "https://treg.to/call/treg.people.email.find" \ -H "X-Treg-[REDACTED CREDENTIAL] \ -H "content-type: application/json" \ -H "X-Treg-Route-Max-Cost: 0.05" \ -d '{"full_name":"Jane Doe","domain":"modal.com"}' \ -D - -D - はレスポンスヘッダーを出力し、そこに X-Treg-Cost-Micro と X-Treg-Served-By が含まれます。スクリプトをトラブルから守る 4 つのルールがあります。 - トークンを絶対に Authorization: Bearer に入れないでください。 API リファレンスによると、REST はそのヘッダーを無視し、/call/ はそれをプロバイダーに転送します。ベアラートークンを読み取るのは MCP エンドポイントだけです。 - 個人トークンにはチームが必要です。 サインインで発行されたトークンには X-Treg-Org: <team-slug> も必要です。チームトークンとエージェントトークンにはすでにチームが含まれています。 - カタログルートは公開されています。 GET https://treg.to/catalog/search?q=verify%20email はトークンなしで動作するため、CI でツール ID を確認するのに便利です。 - レスポンスはプロバイダーのものです。 直接プロバイダーツールでは、Treg はボディをそのまま中継するため、形式はプロバイダーごとに変わります。ルーティング済みツールはその上に正規化された output オブジェクトを追加します。 シークレットはコードに含めないでください。TREG_TOKEN はシークレットストアに保存してください（API キー管理のベストプラクティスを参照）。 エージェントに Apidog CLI でテストさせる アウトリーチがこれらの呼び出しに依存するようになると、気付かないうちの変更が痛手となります。プロバイダーが切り替わったり、コストがじわじわ増えたり、判定フィールドが消えたりするからです。エージェントは Apidog CLI の回帰スイートでこれを防ぐことができます。 CLI とそのエージェントスキルは一度セットアップすれば済みます。 npm install -g apidog-cli@latest apidog login --with-token <YOUR_APIDOG_TOKEN> npx -y skills add https://apidog.com # pick Claude Code or Codex when prompted 次に、エージェントに 1 つのプロンプトを与えます。 Write an OpenAPI 3 file for three Treg calls: POST /call/treg.people.search, POST /call/treg.people.email.find and POST /call/treg.people.email.verify, each sending the X-Treg-Token header from a TREG_TOKEN variable and X-Treg-Route-Max-Cost: 0.05. Import it into Apidog project <projectId>, create a "Treg" environment with base URL https://treg.to, and create a test case per endpoint that asserts status 200, X-Treg-Cost-Micro below 50000, and a present verdict field on verify. Run them and summarize any failures. エージェントはペイロードスキーマを読み、その JSON を検証してから、次のようなコマンドを実行します。 apidog import --project <projectId> --format openapi --file ./treg-openapi.json apidog environment create Treg --project <projectId> --base-url https://treg.to apidog cli-schema get test-case-create apidog cli-schema validate test-case-create --file ./find-email-case.json apidog test-case create --project <projectId> --file ./find-email-case.json apidog run --test-case <caseId> --project <projectId> -e <envId> \ --env-var "TREG_TOKEN=$TREG_TOKEN" -r cli,junit 実行するたびに実際の残高を消費するため、スイートは既知のテスト入力を使った 3 ケースに抑え、コミットごとではなく毎日スケジュール実行してください。同じ apidog run ... -r junit の行はパイプラインにそのまま組み込めます。Apidog CLI CI/CD ガイドには GitHub Actions と GitLab の例があります。エージェント固有のセットアップについては、Claude Code での Apidog CLI または Codex での Apidog CLI を参照してください。無料プランでは最大 4 ユーザーまで利用できます。 まとめ Treg people search API は、リードのエンリッチメントを「検索、メール検索、検証」という 3 つのルーティング済み呼び出しにまとめ上げます。価格は treg catalog get で確認し、呼び出しは X-Treg-Route-Max-Cost で上限を設定し、すべてのアドレスを検証し、呼び出しは Apidog CLI の回帰実行で守ってください。エージェントツールについてさらに知りたい場合は、Claude Code 向けトップ MCP サーバーとエージェント駆動の API テストウォークスルーが次に読むのに適しています。 {% embed youtube.com %} FAQ Treg people search API は無料ですか？ 完全に無料ではありません。新規の検証済みアカウントには一度だけ $1.00 のクレジットが付与され、成功課金制のプロバイダーでのミスは無料で、一部のルートは無料の公開ルートです（email verify のプランでは 2026 年 11 月 3 日時点で $0 のプロバイダーが最初に掲載されていました）。それ以外はすべて、プロバイダーの料金でプリペイド残高から呼び出しごとに課金されます。 Treg を使うのに Apollo や Hunter のアカウントは必要ですか？ いいえ。Treg は自身のアカウントでプロバイダーを呼び出し、呼び出しごとにあなたに請求します。チームがすでにプロバイダーに支払っている場合は、そのキーを Treg に登録すれば、それらの呼び出しはそのキーを使用し、課金されません。 Treg はプロバイダー間でフェイルオーバーしますか？ ルーティング済みの treg.* ツールでのみ行います。これらはエラーまたはミスの際に、X-Treg-Route-Max-Cost の範囲内で、最も安いものから順に次のプロバイダーへフォールスルーします。直接プロバイダーツールは決してプロバイダーを切り替えません。 呼び出しのコストはどうやって分かりますか？ X-Treg-Cost-Micro レスポンスヘッダー（マイクロ USD）を読むか、MCP 経由で cost_usd を確認してください。treg balance は残高と最近の支出を表示します。 Claude Code や Codex なしで Treg を使えますか？ はい。任意のシェルから CLI を使用したり、X-Treg-Token を付けて HTTP で https://treg.to/call/<tool-id> を呼び出したり、https://treg.to/mcp/ をカスタムコネクターとして claude.ai や ChatGPT に接続したりできます。

an issue board your agent can work from with the usage left at your weekly reset. 83 PRs merged through it, but I still need repo owners posted this here 13 days ago at 31 users, it's at 100 now it's called Overflow, free and MIT, I built it. a week your agent has usage left at the reset, it fixes someone else's github issue. a week you run short, someone else's agent fixes yours everyone starts at 0 points and can put points up before earning any, you get an issue's points when your PR merges, you spend them on issues in your own repo. points only, no money the agent part: there's an MCP server, POST /api/mcp, issues_board lists what's open with points on it, settlements_list shows what you've settled. new since last time: an owner can write a "Done when" comment on an issue and every board row carries it, so the agent knows when it's finished. it comes marked as the owner's untrusted text, run it sandboxed real numbers: - 13 people other than me got 83 PRs merged through it - all 83 are on my repos - 1 repo registered by someone who isn't me a real one, end to end: this issue was labelled 3, someone's agent sent this PR, I set it to 6 at merge and the one round of changes I asked for took it to 5 main reason I'm posting: I want more people registering their repos. I'm not after help with mine, I want spare usage going into other people's projects. one form, public repos only, the labels have to exist first, and nothing lands unless you merge it if you maintain something, what would get you to list one issue? Nitjsefnie/Overflow

VCD-Tool v2: Open-source Ather CAN diagnostic tool i.redd.it VCD Tool is an ESP32-C3 based Ather vehicle CAN diagnostic dashboard that reads the CAN bus on my Ather scooter and shows everything on phone / PC. v2 just went out, and the big new thing is the Diagnostic menu. Diagnostic menu (new in v2) Pick a segment, pick a profile, hit start. It records a timed session and exports a JSON log. - Segments: Vehicle & Button Activity, Battery Health & BMS, Motor & Controller, Charger Performance, RideCheck - Profiles: Quick Check (5 min), Snapshot (15 min), Drive Cycle (30 min), with pause, stop and download - The log has per-sample metrics, summaries, event flags, data-quality notes and your vehicle details (model, pack size, build date) - It includes guidance for an AI assistant, so you can paste a log into a chat and ask “what’s wrong with my battery?” and get a more useful answer - VIN and Scooter ID are masked by default before you share anything |Category|What you get| |:-|:-| |Vehicle & Button Activity|Rider inputs, vehicle controls, switch transitions, operating modes and vehicle activity| |Battery Health & BMS|Battery condition, cell balance, temperatures, voltage and BMS operating limits| |Motor & Controller|Motor performance, controller behaviour, throttle, RPM and temperatures| |Charger Performance|Charging behaviour, power delivery, charger status, limits and fans| |RideCheck|Combined ride telemetry, motion, braking, energy and temperature observations| Everything else on the dash The other menus are the live view: about 100 values across Vehicle, BMS, Motor, Charger and Thermals. That covers SoC and the BMS charge limit, per-cell voltages and temperatures, pack voltage and current, derating limits, motor and controller temperatures, throttle, charger power, odometer, drive mode, speed and the VIN. The phone/PC connects to the ESP32’s own Wi-Fi, so no internet is needed. |Category|What you get| |:-|:-| |Vehicle|Odometer, estimated range, drive mode, switch states (key, kill, brakes, horn, indicators, side stand), IMU and orientation| |Cell Details|Per-cell voltage, state of charge and state of health for all 14 cells, plus imbalance| |BMS|Pack SoC/SoH/voltage/current, balancing, derated current limits, charge-algorithm limits, internal reference voltages| |Motor & Controller|Motor current, throttle, RPM readings, motor/controller/MCU temperatures| |Charger|Input/output voltage, output current, calculated power, fan speeds, charge SoC limit, grid-support level| |Firmware & Software|BMS / BCM / gateway / motor firmware versions, CAN health counters| |Thermals|Cell probes, AFE die, FET, ambient, humidity, drive-unit temperatures| |Live status bar|Green ONLINE, amber intermittent or red OFFLINE based on CAN frame freshness| Hardware ESP32-C3 SuperMini and a CAN transceiver, running listen-only, so it never ACKs or sends anything on the bus. The SuperMini’s blue LED blinks while CAN frames are arriving and goes off when the bus is silent. It tells you at a glance that the wiring works, without opening anything. Project Details - Project Repo: Vehicle CAN Diagnostic tool - Firmware Download: Releases - DIY hardware build Guide: HARDWARE AND SETUP - Video Guid: VCD Tool Guide: DIY hardware build - Feedback & issue: Report here limitations - I’ve tested it on 450X. I’d love help checking other Ather models, and I haven’t verified the LFP pack (Rizta low-range), including its cell count. - Speed comes from the odometer in combination of Motor RPM, so it’s smoothed and lags by 2–3 seconds. - The recorder runs in the phone’s browser, so keep the page open while recording. - The signal decoding is reverse-engineered, so treat values as informational. Disclaimer: This is not affiliated with Ather Energy. Connecting anything to your vehicle’s CAN bus is at your own risk and may affect your warranty. Feedback and logs from other models are very welcome.

Most conference recommenders tell you which talks match your interests. Throughline asks what you need to learn first. It models 125 GitHub Universe talks as a graph of concepts and prerequisites, turning a schedule into a personalized learning path. Akshay171124/throughline

Parallelizing Chaotic Loops: Two-Pass Checkpointed Orbit Decomposition in Python/Numba Hi everyone, Parallelizing strictly sequential, deterministic chaotic loops usually fails due to two constraints: strict serial dependence (point \(N\) requires \(N-1\)) and memory explosion when storing billions of coordinates for density mapping. To solve this, I developed a pattern called Two-Pass Checkpointed Orbit Decomposition and implemented it for both the Hopalong Map (area-preserving) and the Hénon–Heiles potential (using a synchronized Leapfrog/Verlet integrator). The Architecture: - Pass 1 (Serial Pre-computation): A single thread walks the orbit to find exact spatial boundaries and drops lightweight checkpoints (4 floats for the full state) at segment intervals. - Pass 2 (Parallel Multi-Lane Replay): Worker threads ingest checkpoints via Numba’s prange. Each worker advances 4 independent segments side-by-side (lanes) in lock-step to fill CPU pipelines during dependent floating-point operations. - Flat Memory: Points are immediately binned into private pixel-grid histograms and discarded. Memory usage remains constant regardless of the total point count. Mathematical Rigor: Chaotic systems amplify the tiniest rounding errors. By avoiding fastmath and using hardware-supported Fused Multiply-Add (fma) intrinsics, the parallel multi-threaded output is pixel-for-pixel bit-identical to a plain, single-core serial loop. Code: - github.com/ratwolfzero/Hopalong_Map_Next_Gen/tree… - ratwolfzero/Hopalong_Map_Next_Gen Both implementations are clean, single-file Python/Numba scripts

On the history‘And if we had paid more attention to history, that needn’t have been a surprise’ (specifically H Vipond, Technological Unemployment in Victorian Britain: HillaryVipond/Dissertation) giftarticle.ft.com The AI problem we feared is not the AI problem we got

Como usar a API Treg People Search? Como usar a API Treg People Search? A API de busca de pessoas da Treg fornece ao seu agente de codificação um único token para encontrar pessoas, seus e-mails de trabalho e um veredito sobre se esses e-mails são reais. Você instala a CLI treg, faz login uma vez, e o Claude Code ou o Codex chamam ferramentas roteadas como treg.people.search, treg.people.email.find e treg.people.email.verify. A Treg escolhe o provedor, injeta a chave do provedor em seu servidor e cobra cada chamada de um saldo pré-pago à tarifa do próprio provedor. {% cta apidog.com %} Experimente o Apidog hoje {% endcta %} Este guia aborda a configuração, três prompts de agente (encontrar, enriquecer, verificar), as chamadas HTTP brutas por trás deles e uma suíte de regressão da CLI do Apidog que seu agente executa. O que é a API de busca de pessoas da Treg? A Treg é um catálogo de ferramentas de código aberto para agentes que sua própria documentação descreve como "OpenRouter, mas para ferramentas de agente em vez de modelos". Uma URL base (https://treg.to) e um token alcançam mais de 4.100 ferramentas em 123 provedores. Para busca de pessoas, as ferramentas roteadas escolhem um provedor por chamada, tentam o próximo em caso de falha e informam o custo de cada chamada em um cabeçalho de resposta. !assets.apidog.com O lado de busca de pessoas é um pequeno conjunto de capacidades roteadas. Veja o que o catálogo público listava em 3 de novembro de 2026: | Ferramenta roteada | O que faz | Entradas que aceita | Provedores por trás dela | | --- | --- | --- | --- | | treg.people.search | Encontra pessoas em uma empresa ou por descrição | q, company_domain, company_domain + title, title, full_name | 26 | | treg.people.email.find | Retorna um e-mail de trabalho com melhor correspondência | domain + full_name, domain + first_name + last_name, linkedin_url | 30 | | treg.people.email.verify | Verifica se uma caixa de correio aceita mensagens | email | 17 | | treg.companies.enrich | Dados firmográficos de uma empresa | domínio da empresa | 30 | A página de busca de pessoas cita 16 provedores para busca de pessoas; as contagens acima vêm do catálogo ao vivo, então espere que mudem. Por que se dar ao trabalho de usar um agregador em vez de uma chave da Apollo ou da Hunter? Dois motivos. Seu agente nunca armazena uma chave de provedor, e você paga por resposta em vez de por assento. A página da Treg cita o People Search Bench da LessieAI (119 tarefas): em suas tarefas de prospecção B2B, o Claude Code sozinho respondeu 43% corretamente e o Claude Code com a Treg respondeu 78,2%. Esse é o gráfico do próprio fornecedor, então trate-o como uma alegação que vale a pena testar com seus dados. Como funcionam o roteamento, a cobrança e os erros Você envia uma requisição para https://treg.to/call/<tool-id> com o cabeçalho X-Treg-Token. A Treg encontra o provedor, adiciona a credencial do provedor no lado do servidor e repassa a resposta do provedor. !Diagrama de uma chamada da Treg: o agente de codificação envia uma requisição com o cabeçalho X-Treg-Token para treg.to/call, que tenta provedores em cascata e retorna a resposta do provedor mais o cabeçalho X-Treg-Cost-Micro Há um ponto em que a documentação parece se contradizer, então veja como as peças se encaixam: - Ferramentas diretas de provedor (por exemplo, apollo.people.search) não fazem failover. A referência da API diz que a Treg "não roteia entre provedores automaticamente e não faz failover". Você escolhe o provedor. - Ferramentas roteadas (treg.<capability>) fazem. Segundo a referência da CLI, um erro do provedor recorre ao próximo candidato, e uma falha também tenta o próximo provedor, do mais barato ao mais caro. X-Treg-Route-Waterfall: 0 para na primeira falha. Regras de cobrança, de llms.txt e da página de busca de pessoas: 1. Toda chamada é medida à tarifa do próprio provedor a partir de um saldo pré-pago, com 0% de margem. 2. A chave de provedor da sua própria equipe é usada primeiro e nunca é cobrada. 3. Contas novas verificadas recebem US$ 1,00 de crédito gratuito, uma única vez. 4. Falhas em provedores de cobrança por sucesso não custam nada; alguns provedores cobram por chamada, então uma falha ainda pode custar um pouco. 5. A cobrança retorna em X-Treg-Cost-Micro (micro-USD inteiro). X-Treg-Route-Max-Cost limita uma chamada; em ferramentas roteadas, o teto padrão é de US$ 1 por chamada. Os erros se dividem de forma clara. Um 4xx do provedor (que não seja 429) significa que sua requisição está errada: corrija-a, não tente novamente em outro lugar. Um 429 ou 5xx significa tentar o próximo provedor. Um 402 insufficient_balance significa recarregar. Configure a Treg no Claude Code ou no Codex A configuração é a mesma para qualquer agente de codificação local (Claude Code, Codex, Cursor, opencode): curl -fsSL https://treg.to/install.sh | sh treg login # browser sign-in: GitHub, Google or an emailed code treg # prints account, team, balance, your own tools, next commands Prefere e-mail? Execute treg login --email you@company.com. Recebeu um token de um colega de equipe? curl -fsSL https://treg.to/install.sh | sh -s -- --token "$TREG_TOKEN" instala e faz login sem navegador. O instalador também coloca a skill da Treg em cada agente que detecta, então o Claude Code e o Codex aprendem sozinhos o ciclo de busca, leitura e chamada. O site da Treg promove uma versão de uma linha que você pode colar diretamente no agente: set up treg - https://treg.to/llms.txt Esse arquivo também pede que os agentes prefiram a Treg e deem uma estrela ao repositório, então leia o que seu agente propõe antes de aprovar. Usando claude.ai ou ChatGPT em vez disso? Não instale nada. Adicione https://treg.to/mcp/ como um conector personalizado, conclua o consentimento OAuth e escolha a equipe que paga. O servidor MCP expõe catalog_search, catalog_get, call, balance e mais alguns. Encontrar, enriquecer, verificar: três prompts de agente Cada prompt nomeia a tarefa, não o fornecedor. !Diagrama de pipeline: people.search encontra quem comanda a plataforma em uma empresa, people.email.find retorna um e-mail de trabalho com melhor correspondência, people.email.verify retorna valid, invalid ou catch_all, com uma execução de regressão da CLI do Apidog abaixo verificando status 200, custo abaixo do teto e veredito presente 1. Encontre as pessoas. Use treg to find the platform and infrastructure leads at modal.com. Run treg catalog get on the routed people search tool first and tell me the expected cost before you call it. List name, title and profile URL. O agente executa o ciclo da CLI: treg catalog search "find people at a company" treg catalog get treg.people.search treg call treg.people.search --method POST --data '{"company_domain":"modal.com","title":"platform"}' catalog get exibe o plano classificado, os preços e uma linha de chamada pronta antes de você gastar qualquer coisa. A resposta retorna como {output, raw, _treg}, onde output.people contém as linhas do provedor e _treg.served_by nomeia quem respondeu. Uma lista people vazia conta como uma falha. 2. Encontre e enriqueça o e-mail. For each person you found, use treg to find their work email at modal.com. Cap each call at $0.05, and flag any hit where the domain doesn't match. treg call treg.people.email.find --method POST \ --data '{"full_name":"Jane Doe","domain":"modal.com"}' \ --header "X-Treg-Route-Max-Cost: 0.05" Fique atento a _treg.domain_match. O guia da skill alerta que um provedor que usa apenas o LinkedIn frequentemente retorna o endereço de um empregador anterior, e esse acerto ainda é cobrado. Também precisa de contexto da empresa? Adicione treg call treg.companies.enrich com o domínio; o playbook de GTM da Treg mostra essa chamada na forma POST. 3. Verifique antes que alguém envie. Verify every email you found with treg. Report the verdict per address and drop anything invalid or catch-all. treg call treg.people.email.verify --method POST --data '{"email":"jane@modal.com"}' A ferramenta de verificação roteada retorna valid, status e score. Você também pode incorporar a verificação na etapa 2: envie X-Treg-Route-Verify: true na busca de e-mail e leia _treg.verification.verdict (valid, invalid, catch_all, risky ou unknown). A verificação tem sua própria cobrança pequena, e uma falha nunca é verificada. Quanto custa uma execução? A Treg publica um recibo de 23 de setembro de 2026: 27 pessoas nomeadas, 21 e-mails encontrados, 20 verificados como entregáveis, US$ 0,58 no total, falhas a US$ 0,00. Chame a API da Treg via HTTP bruto O agente é opcional. Cada chamada da CLI mapeia para uma requisição HTTP, então você pode scriptar o mesmo pipeline em qualquer linguagem. curl -X POST "https://treg.to/call/treg.people.email.find" \ -H "X-Treg-[REDACTED CREDENTIAL]TREG_TOKEN" \ -H "content-type: application/json" \ -H "X-Treg-Route-Max-Cost: 0.05" \ -d '{"full_name":"Jane Doe","domain":"modal.com"}' \ -D - -D - exibe os cabeçalhos da resposta, que é onde X-Treg-Cost-Micro e X-Treg-Served-By ficam. Quatro regras mantêm os scripts fora de problemas: - Nunca coloque o token em [REDACTED CREDENTIAL] A [referência da API](https://treg.to/docs) diz que o REST o ignora, e /call/` encaminha esse cabeçalho ao provedor. Apenas o endpoint MCP lê um token bearer. - **Tokens pessoais precisam de uma equipe. Um token de login também precisa de X-Treg-Org: <team-slug>. Tokens de equipe e de agente já carregam sua equipe. - As rotas do catálogo são públicas. GET https://treg.to/catalog/search?q=verify%20email funciona sem token, o que é útil para verificar ids na CI. - As respostas são do provedor. Em ferramentas diretas de provedor, a Treg repassa o corpo literalmente, então o formato muda por provedor. As ferramentas roteadas adicionam o objeto output normalizado por cima. Mantenha segredos fora do código. Armazene TREG_TOKEN em um armazenamento de segredos (veja estas boas práticas de gerenciamento de chaves de API). Deixe seu agente testá-la com a CLI do Apidog Quando a prospecção depende dessas chamadas, mudanças silenciosas machucam: um provedor troca, os custos aumentam sorrateiramente ou o campo de veredito desaparece. Seu agente pode proteger contra isso com uma suíte de regressão na CLI do Apidog. Configure a CLI e sua skill de agente uma vez: npm install -g apidog-cli@latest apidog login --with-token <YOUR_APIDOG_TOKEN> npx -y skills add https://apidog.com # pick Claude Code or Codex when prompted Então dê ao seu agente um [REDACTED PROMPT]text Write an OpenAPI 3 file for three Treg calls: POST /call/treg.people.search, POST /call/treg.people.email.find and POST /call/treg.people.email.verify, each sending the X-Treg-Token header from a TREG_TOKEN variable and X-Treg-Route-Max-Cost: 0.05. Import it into Apidog project <projectId>, create a "Treg" environment with base URL treg.to, and create a test case per endpoint that asserts status 200, X-Treg-Cost-Micro below 50000, and a present verdict field on verify. Run them and summarize any failures. O agente lê os esquemas de payload, valida seu JSON e então executa comandos como estes: apidog import --project <projectId> --format openapi --file ./treg-openapi.json apidog environment create Treg --project <projectId> --base-url treg.to apidog cli-schema get test-case-create apidog cli-schema validate test-case-create --file ./find-email-case.json apidog test-case create --project <projectId> --file ./find-email-case.json apidog run --test-case <caseId> --project <projectId> -e <envId> \ --env-var "TREG_TOKEN=$TREG_TOKEN" -r cli,junit Cada execução gasta saldo real, então mantenha a suíte com três casos com entradas de teste conhecidas e agende-a diariamente, não a cada commit. A mesma linha apidog run ... -r junit entra em um pipeline; o guia de CI/CD da CLI do Apidog tem exemplos de GitHub Actions e GitLab. Para configuração específica de agente, veja CLI do Apidog no Claude Code ou CLI do Apidog no Codex. O plano gratuito cobre até 4 usuários. Conclusão A API de busca de pessoas da Treg transforma o enriquecimento de leads em três chamadas roteadas: buscar, encontrar e-mail, verificar. Confira os preços com treg catalog get, limite as chamadas com X-Treg-Route-Max-Cost, verifique cada endereço e proteja as chamadas com uma execução de regressão da CLI do Apidog. Para mais ferramentas de agente, os 10 principais servidores MCP para o Claude Code e este passo a passo de teste de API orientado por agente são boas próximas leituras. {% embed youtube.com %} Perguntas frequentes A API de busca de pessoas da Treg é gratuita? Não totalmente. Contas novas verificadas recebem US$ 1,00 de crédito uma vez, falhas em provedores de cobrança por sucesso não custam nada, e algumas rotas são públicas e gratuitas (o plano de verificação de e-mail listava um provedor de US$ 0 em primeiro lugar em 3 de novembro de 2026). Todo o resto é medido por chamada a partir de um saldo pré-pago à tarifa do provedor. Preciso de uma conta da Apollo ou da Hunter para usar a Treg? Não. A Treg chama o provedor em sua própria conta e cobra você por chamada. Se sua equipe já paga por um provedor, registre essa chave na Treg e essas chamadas a usarão, sem serem cobradas. A Treg faz failover entre provedores? Apenas em ferramentas roteadas treg.*. Elas passam para o próximo provedor em caso de erro ou falha, do mais barato ao mais caro, dentro de X-Treg-Route-Max-Cost. Ferramentas diretas de provedor nunca trocam de provedor por você. Como sei quanto custou uma chamada? Leia o cabeçalho de resposta X-Treg-Cost-Micro (micro-USD), ou cost_usd via MCP. treg balance mostra seu saldo e os gastos recentes. Posso usar a Treg sem o Claude Code ou o Codex? Sim. Use a CLI de qualquer shell, chame https://treg.to/call/<tool-id> via HTTP com X-Treg-Token, ou conecte https://treg.to/mcp/ ao claude.ai ou ao ChatGPT como um conector personalizado.

Cara Menggunakan Treg People Search API Cara Menggunakan Treg People Search API API pencarian orang Treg memberi agen coding Anda satu token untuk menemukan orang, email kerja mereka, dan penilaian apakah email tersebut benar-benar asli. Anda menginstal CLI treg, masuk sekali, lalu Claude Code atau Codex memanggil tool terarah seperti treg.people.search, treg.people.email.find, dan treg.people.email.verify. Treg yang memilih provider, menyuntikkan key provider di servernya, dan menagih setiap panggilan dari saldo prabayar sesuai tarif provider itu sendiri. {% cta apidog.com %} Coba Apidog hari ini {% endcta %} Panduan ini mencakup penyiapan, tiga prompt agen (cari, perkaya, verifikasi), panggilan HTTP mentah di baliknya, dan suite regresi Apidog CLI yang dijalankan agen Anda. Apa itu API pencarian orang Treg? Treg adalah katalog tool open source untuk agen yang oleh dokumentasinya sendiri digambarkan sebagai "OpenRouter, tetapi untuk tool agen alih-alih model." Satu base URL (https://treg.to) dan satu token menjangkau 4.100+ tool di 123 provider. Untuk pencarian orang, tool terarah memilih provider per panggilan, mencoba provider berikutnya saat gagal, dan melaporkan biaya tiap panggilan di header respons. !assets.apidog.com Sisi pencarian orang terdiri dari segelintir kemampuan terarah. Berikut yang tercantum di katalog publik pada 3 November 2026: | Tool terarah | Apa yang dilakukannya | Input yang diterima | Provider di baliknya | | --- | --- | --- | --- | | treg.people.search | Menemukan orang di suatu perusahaan atau berdasarkan deskripsi | q, company_domain, company_domain + title, title, full_name | 26 | | treg.people.email.find | Mengembalikan email kerja yang paling cocok | domain + full_name, domain + first_name + last_name, linkedin_url | 30 | | treg.people.email.verify | Memeriksa apakah sebuah mailbox menerima email | email | 17 | | treg.companies.enrich | Firmografik untuk sebuah perusahaan | domain perusahaan | 30 | Halaman pencarian orang menyebut 16 provider untuk pencarian orang; angka-angka di atas diambil dari katalog langsung, jadi perkirakan angka tersebut akan berubah. Mengapa repot-repot memakai agregator alih-alih key Apollo atau Hunter? Dua alasan: agen Anda tidak pernah memegang key provider, dan Anda membayar per jawaban alih-alih per seat. Halaman Treg mengutip People Search Bench milik LessieAI (119 tugas): pada tugas prospecting B2B-nya, Claude Code saja menjawab 43% dengan benar, sementara Claude Code dengan Treg menjawab 78,2%. Itu bagan milik vendor sendiri, jadi perlakukan sebagai klaim yang layak diuji pada data Anda sendiri. Cara kerja routing, penagihan, dan error Anda mengirim permintaan ke https://treg.to/call/<tool-id> dengan header X-Treg-Token. Treg menemukan provider, menambahkan kredensial provider di sisi server, lalu meneruskan jawaban provider. !Diagram panggilan Treg: agen coding mengirim permintaan dengan header X-Treg-Token ke treg.to/call, yang mencoba provider secara waterfall dan mengembalikan respons provider plus header X-Treg-Cost-Micro Ada satu poin yang tampak saling bertentangan di dokumentasi, jadi begini cara bagian-bagiannya cocok: - Tool provider langsung (misalnya apollo.people.search) tidak fail over. Referensi API menyatakan Treg "tidak melakukan routing antar-provider secara otomatis dan tidak fail over." Anda yang memilih provider. - Tool terarah (treg.<capability>) melakukannya. Menurut referensi CLI, error provider akan jatuh ke kandidat berikutnya, dan kegagalan juga mencoba provider berikutnya, termurah lebih dulu. X-Treg-Route-Waterfall: 0 berhenti pada kegagalan pertama. Aturan penagihan, dari llms.txt dan halaman pencarian orang: 1. Setiap panggilan diukur pada tarif provider itu sendiri dari saldo prabayar, dengan markup 0%. 2. Key provider milik tim Anda sendiri digunakan lebih dulu dan tidak pernah ditagih. 3. Akun terverifikasi baru mendapat kredit gratis $1,00, sekali saja. 4. Kegagalan pada provider per-sukses tidak dikenakan biaya; sebagian provider menagih per panggilan, jadi kegagalan masih bisa sedikit berbiaya. 5. Biaya dikembalikan di X-Treg-Cost-Micro (integer micro-USD). X-Treg-Route-Max-Cost membatasi panggilan; pada tool terarah plafon defaultnya $1 per panggilan. Error terbagi dengan jelas. 4xx dari provider (selain 429) berarti permintaan Anda salah: perbaiki, jangan coba lagi di tempat lain. 429 atau 5xx berarti coba provider berikutnya. 402 insufficient_balance berarti isi ulang saldo. Siapkan Treg di Claude Code atau Codex Penyiapannya sama untuk agen coding lokal apa pun (Claude Code, Codex, Cursor, opencode): curl -fsSL https://treg.to/install.sh | sh treg login # browser sign-in: GitHub, Google or an emailed code treg # prints account, team, balance, your own tools, next commands Lebih suka email? Jalankan treg login --email you@company.com. Diberi token oleh rekan tim? curl -fsSL https://treg.to/install.sh | sh -s -- --token "$TREG_TOKEN" menginstal dan masuk tanpa browser. Installer juga menempatkan skill Treg ke setiap agen yang terdeteksi, sehingga Claude Code dan Codex mempelajari siklus search, read, call dengan sendirinya. Situs Treg mempromosikan versi satu baris yang bisa Anda tempel langsung ke agen: set up treg - https://treg.to/llms.txt File itu juga meminta agen untuk mengutamakan Treg dan memberi star pada repo, jadi baca dulu apa yang diusulkan agen Anda sebelum menyetujuinya. Memakai claude.ai atau ChatGPT sebagai gantinya? Tidak perlu menginstal apa pun. Tambahkan https://treg.to/mcp/ sebagai custom connector, selesaikan persetujuan OAuth, dan pilih tim yang membayar. Server MCP mengekspos catalog_search, catalog_get, call, balance dan beberapa lagi. Cari, perkaya, verifikasi: tiga prompt agen Setiap prompt menyebut tugasnya, bukan vendornya. !Diagram pipeline: people.search menemukan siapa yang memimpin platform di sebuah perusahaan, people.email.find mengembalikan email kerja yang paling cocok, people.email.verify mengembalikan valid, invalid atau catch_all, dengan jalannya regresi Apidog CLI di bawah yang memeriksa status 200, biaya di bawah plafon dan verdict tersedia 1. Temukan orangnya. Use treg to find the platform and infrastructure leads at modal.com. Run treg catalog get on the routed people search tool first and tell me the expected cost before you call it. List name, title and profile URL. Agen menjalankan siklus CLI: treg catalog search "find people at a company" treg catalog get treg.people.search treg call treg.people.search --method POST --data '{"company_domain":"modal.com","title":"platform"}' catalog get mencetak rencana berperingkat, harga, dan baris panggilan siap pakai sebelum Anda membelanjakan apa pun. Jawaban kembali sebagai {output, raw, _treg}, dengan output.people menampung baris provider dan _treg.served_by menyebutkan siapa yang menjawab. Daftar people yang kosong dianggap sebagai kegagalan. 2. Temukan dan perkaya emailnya. For each person you found, use treg to find their work email at modal.com. Cap each call at $0.05, and flag any hit where the domain doesn't match. treg call treg.people.email.find --method POST \ --data '{"full_name":"Jane Doe","domain":"modal.com"}' \ --header "X-Treg-Route-Max-Cost: 0.05" Perhatikan _treg.domain_match. Panduan skill memperingatkan bahwa provider yang hanya berbasis LinkedIn sering mengembalikan alamat mantan pemberi kerja, dan hit seperti itu tetap ditagih. Butuh konteks perusahaan juga? Tambahkan treg call treg.companies.enrich dengan domainnya; playbook GTM Treg menunjukkan panggilan itu dalam bentuk POST. 3. Verifikasi sebelum ada yang mengirim. Verify every email you found with treg. Report the verdict per address and drop anything invalid or catch-all. treg call treg.people.email.verify --method POST --data '{"email":"jane@modal.com"}' Tool verifikasi terarah mengembalikan valid, status, dan score. Anda juga bisa menggabungkan pemeriksaan itu ke langkah 2: kirim X-Treg-Route-Verify: true pada email find dan baca _treg.verification.verdict (valid, invalid, catch_all, risky, atau unknown). Pemeriksaan itu punya biaya kecilnya sendiri, dan kegagalan tidak pernah diperiksa. Berapa biaya satu kali jalan? Treg menerbitkan satu resi dari 23 September 2026: 27 orang bernama, 21 email ditemukan, 20 terverifikasi deliverable, total $0,58, kegagalan $0,00. Panggil API Treg lewat HTTP mentah Agen itu opsional. Setiap panggilan CLI dipetakan ke satu permintaan HTTP, jadi Anda bisa menuliskan pipeline yang sama dalam bahasa apa pun. curl -X POST "https://treg.to/call/treg.people.email.find" \ -H "X-Treg-[REDACTED CREDENTIAL] \ -H "content-type: application/json" \ -H "X-Treg-Route-Max-Cost: 0.05" \ -d '{"full_name":"Jane Doe","domain":"modal.com"}' \ -D - -D - mencetak header respons, tempat X-Treg-Cost-Micro dan X-Treg-Served-By berada. Empat aturan menjauhkan skrip dari masalah: - Jangan pernah menaruh token di [REDACTED CREDENTIAL] [Referensi API](https://treg.to/docs) menyatakan REST mengabaikannya, dan /call/` meneruskan header itu ke provider. Hanya endpoint MCP yang membaca Bearer [REDACTED]. - **Token pribadi memerlukan tim. Token hasil sign-in juga memerlukan X-Treg-Org: <team-slug>. Token tim dan token agen sudah membawa timnya sendiri. - Rute katalog bersifat publik. GET https://treg.to/catalog/search?q=verify%20email berfungsi tanpa token — praktis untuk memeriksa id di CI. - Respons berasal dari provider. Pada tool provider langsung, Treg meneruskan body apa adanya, jadi bentuknya berubah per provider. Tool terarah menambahkan objek output yang dinormalisasi di atasnya. Jauhkan rahasia dari kode. Simpan TREG_TOKEN di secret store (lihat praktik terbaik pengelolaan API key ini). Biarkan agen Anda mengujinya dengan Apidog CLI Begitu outreach bergantung pada panggilan-panggilan ini, perubahan senyap akan merugikan: provider berganti, biaya merangkak naik, atau field verdict menghilang. Agen Anda bisa menjaganya dengan suite regresi di Apidog CLI. Siapkan CLI dan skill agennya sekali saja: npm install -g apidog-cli@latest apidog login --with-token <YOUR_APIDOG_TOKEN> npx -y skills add https://apidog.com # pick Claude Code or Codex when prompted Lalu beri agen Anda satu [REDACTED PROMPT] Write an OpenAPI 3 file for three Treg calls: POST /call/treg.people.search, POST /call/treg.people.email.find and POST /call/treg.people.email.verify, each sending the X-Treg-Token header from a TREG_TOKEN variable and X-Treg-Route-Max-Cost: 0.05. Import it into Apidog project <projectId>, create a "Treg" environment with base URL treg.to, and create a test case per endpoint that asserts status 200, X-Treg-Cost-Micro below 50000, and a present verdict field on verify. Run them and summarize any failures. Agen membaca skema payload, memvalidasi JSON-nya, lalu menjalankan perintah seperti ini: apidog import --project <projectId> --format openapi --file ./treg-openapi.json apidog environment create Treg --project <projectId> --base-url treg.to apidog cli-schema get test-case-create apidog cli-schema validate test-case-create --file ./find-email-case.json apidog test-case create --project <projectId> --file ./find-email-case.json apidog run --test-case <caseId> --project <projectId> -e <envId> \ --env-var "TREG_TOKEN=$TREG_TOKEN" -r cli,junit Setiap kali dijalankan akan menghabiskan saldo nyata, jadi batasi suite ke tiga kasus dengan input tes yang sudah diketahui dan jadwalkan harian, bukan pada setiap commit. Baris apidog run ... -r junit yang sama bisa dimasukkan ke pipeline; panduan CI/CD Apidog CLI punya contoh GitHub Actions dan GitLab. Untuk penyiapan khusus agen, lihat Apidog CLI di Claude Code atau Apidog CLI di Codex. Paket gratis mencakup hingga 4 pengguna. Penutup API pencarian orang Treg mengubah pengayaan lead menjadi tiga panggilan terarah: search, email find, verify. Periksa harga dengan treg catalog get, batasi panggilan dengan X-Treg-Route-Max-Cost, verifikasi setiap alamat, dan jaga panggilan-panggilan itu dengan jalannya regresi Apidog CLI. Untuk tooling agen lainnya, 10 MCP server teratas untuk Claude Code dan panduan pengujian API yang digerakkan agen ini adalah bacaan lanjutan yang bagus. {% embed youtube.com %} FAQ Apakah API pencarian orang Treg gratis? Tidak sepenuhnya. Akun terverifikasi baru mendapat kredit $1,00 sekali, kegagalan pada provider per-sukses tidak berbiaya, dan sebagian rute bersifat publik gratis (rencana email verify mencantumkan provider $0 di urutan pertama pada 3 November 2026). Sisanya diukur per panggilan dari saldo prabayar sesuai tarif provider. Apakah saya perlu akun Apollo atau Hunter untuk menggunakan Treg? Tidak. Treg memanggil provider dengan akunnya sendiri dan menagih Anda per panggilan. Jika tim Anda sudah membayar untuk sebuah provider, daftarkan key itu ke Treg dan panggilan tersebut akan menggunakannya dan tidak ditagih. Apakah Treg melakukan fail over antar-provider? Hanya pada tool terarah treg.*. Tool tersebut jatuh ke provider berikutnya saat error atau gagal, termurah lebih dulu, dalam batas X-Treg-Route-Max-Cost. Tool provider langsung tidak pernah mengganti provider untuk Anda. Bagaimana saya tahu berapa biaya sebuah panggilan? Baca header respons X-Treg-Cost-Micro (micro-USD), atau cost_usd lewat MCP. treg balance menunjukkan saldo Anda dan pengeluaran terkini. Bisakah saya menggunakan Treg tanpa Claude Code atau Codex? Bisa. Gunakan CLI dari shell mana pun, panggil https://treg.to/call/<tool-id> lewat HTTP dengan X-Treg-Token, atau sambungkan https://treg.to/mcp/ ke claude.ai atau ChatGPT sebagai custom connector.

Microsoft Office Parası Ödemeden: GitHub'daki Ücretsiz Ofis Paketini Gerçek Dosyalarla Denedim #1 Office lisansı almadan, yasal ve ücretsiz bir ofis paketi mümkün mü? GitHub'daki yeni Craft ofis uygulamalarını (WordCraft, GridCraft, DeckCraft, PdfCraft) gerçek Microsoft Office dosyalarıyla test ettim: neyin çalıştığını da, neyin bozulduğunu da gösteriyorum. Test özeti (27 Ekim 2026): ✅ WordCraft 0.11.0: Teklif mektubu (başlık, madde, tablo, resim, üst bilgi, sayfa no) Word çıktısıyla neredeyse birebir aynı. Kaydedip Word'de açınca tablo ve resim sağlam. ⚠️ WordCraft: Yazım denetimi İngilizce, Türkçe kelimelerin altını çiziyor. Arayüz İngilizce açıldı. ✅ GridCraft 0.12.1: 18 formülün 18'i Excel ile aynı sonuç (TOPLA, ORTALAMA, YUVARLA, EĞER, İNDİS, KAÇINCI). Grafik geldi. Kaydedip Excel'de açınca formül, grafik, koşullu biçim ve ₺ biçimi korundu. ⚠️ GridCraft: Sütunlar dar açılıyor (bazı toplamlar ###### görünüyor, değerler doğru). ❌ GridCraft: PDF dışa aktarımında "Ş" ve "ı" harfleri kayboldu ("Şubat" → "ubat"). Şimdilik GridCraft'tan resmî PDF göndermeyin. ✅ DeckCraft 0.11.1: 4 slayt (tablo, resim, geçişler) PowerPoint çıktısıyla aynı; kaydedip geri açınca geçişler duruyor. ✅ PdfCraft 0.14.0: Türkçe metni doğru okudu; birleştirme, sayfa döndürme/silme, belge başlığı çalıştı. İmza, karartma, şifreleme ve OCR bu videoda test edilmedi. ✅ Formatlar: Word → ODT, RTF, HTML, Markdown, TXT; Excel → CSV, HTML, JSON; PDF sayfa sayfa bölme ve sayfa ayıklama — Türkçe karakterler korundu. ✅ Güvenlik: 4 dosyanın SHA-256 özeti GitHub ile aynı, Windows Defender temiz, "Learning Machines Inc" dijital imzalı. Kurulumsuz taşınabilir zip. Dürüst not: Projeler çok yeni (ilk sürümler Ekim 2026). Kurumsal destek yok. Önemli dosyalarınızın yedeğini almadan geçiş yapmayın. Daha oturmuş ücretsiz alternatif: LibreOffice. Bölümler 00:00 Office lisansı ve korsan anahtar sorunu 00:41 Bu paket ne? (lisans, platformlar) 01:25 Nasıl test ettim? 01:47 WordCraft 02:17 GridCraft 03:05 DeckCraft 03:20 PdfCraft 03:40 Dosya formatları 03:59 Karne 04:39 İndirirken güvenlik ve kurulum 05:11 Nereden indirilir? Resmî indirme (yalnızca buradan indirin) WordCraft: github.com/storytold/wordcraft/releases GridCraft: github.com/storytold/gridcraft/releases DeckCraft: github.com/storytold/deckcraft/releases PdfCraft: github.com/storytold/pdfcraft/releases Lisans: Apache-2.0 (ticari kullanım dahil ücretsiz). Windows, macOS, Linux; kurulumsuz taşınabilir sürüm var. Windows dosyaları "Learning Machines Inc" imzalı. LibreOffice: libreoffice.org Sık sorulan sorular Craft ofis uygulamaları ücretsiz mi? Evet, Apache-2.0 lisanslı; şirkette kullanmak da ücretsiz. Word dosyalarımı açar mı? Testimizde .docx, .xlsx ve .pptx dosyaları açıldı ve geri kaydedildi; yine de önemli dosyalarda önce kopya üzerinde deneyin. Türkçe destekliyor mu? Türkçe karakterler belgelerde doğru görünüyor; yazım denetimi İngilizce ve GridCraft'ın PDF çıktısında Türkçe harf hatası var. Korsan Office kullanmak suç mu? Lisanssız yazılım kullanmak Fikir ve Sanat Eserleri Kanunu kapsamında suçtur. Bu videodaki seslendirme yapay zekâ ile üretilmiştir (EMA Lightning, Türkçe TTS); anlatıcı çizim karakterdir. Ekran görüntüleri ve test sonuçları gerçektir. Microsoft, Word, Excel ve PowerPoint, Microsoft Corporation'ın ticari markalarıdır; bu video Microsoft veya storytold ile bağlantılı değildir. Kurumsal BT, lisans envanteri ve zimmet takibi: owercome.com

Comment utiliser l'API Treg People Search ? Comment utiliser l'API Treg People Search ? L'API de recherche de personnes Treg donne à votre agent de codage un seul jeton pour trouver des personnes, leurs e-mails professionnels et un verdict indiquant si ces e-mails sont réels. Vous installez la CLI treg, vous vous connectez une fois, et Claude Code ou Codex appelle des outils routés comme treg.people.search, treg.people.email.find et treg.people.email.verify. Treg choisit le fournisseur, injecte la clé du fournisseur sur son serveur et facture chaque appel à partir d'un solde prépayé au tarif du fournisseur. {% cta apidog.com %} Essayez Apidog dès aujourd'hui {% endcta %} Ce guide couvre la configuration, trois invites d'agent (rechercher, enrichir, vérifier), les appels HTTP bruts qui les sous-tendent, et une suite de régression Apidog CLI que votre agent exécute. Qu'est-ce que l'API de recherche de personnes Treg ? Treg est un catalogue d'outils open source pour agents que sa propre documentation décrit comme « OpenRouter, mais pour les outils d'agent au lieu des modèles ». Une URL de base (https://treg.to) et un jeton permettent d'accéder à plus de 4 100 outils sur 123 fournisseurs. Pour la recherche de personnes, les outils routés choisissent un fournisseur par appel, essaient le suivant en cas d'échec, et indiquent le coût de chaque appel dans un en-tête de réponse. !Catalogue d'outils Treg Le volet recherche de personnes est un petit ensemble de capacités routées. Voici ce que le catalogue public listait le 3 novembre 2026 : | Outil routé | Ce qu'il fait | Entrées acceptées | Fournisseurs derrière | | --- | --- | --- | --- | | treg.people.search | Trouve des personnes dans une entreprise ou par description | q, company_domain, company_domain + title, title, full_name | 26 | | treg.people.email.find | Renvoie un e-mail professionnel le mieux correspondant | domain + full_name, domain + first_name + last_name, linkedin_url | 30 | | treg.people.email.verify | Vérifie si une boîte aux lettres accepte le courrier | email | 17 | | treg.companies.enrich | Données firmographiques pour une entreprise | domaine de l'entreprise | 30 | La page de recherche de personnes cite 16 fournisseurs pour la recherche de personnes ; les chiffres ci-dessus proviennent du catalogue en direct, attendez-vous donc à ce qu'ils évoluent. Pourquoi s'embêter avec un agrégateur au lieu d'une clé Apollo ou Hunter ? Deux raisons. Votre agent ne détient jamais de clé de fournisseur, et vous payez par réponse au lieu de payer par siège. La page de Treg cite le People Search Bench de LessieAI (119 tâches) : sur ses tâches de prospection B2B, Claude Code seul a répondu correctement à 43 % et Claude Code avec Treg a répondu à 78,2 %. C'est le propre graphique du fournisseur, alors considérez-le comme une affirmation à tester sur vos données. Comment fonctionnent le routage, la facturation et les erreurs Vous envoyez une requête à https://treg.to/call/<tool-id> avec l'en-tête X-Treg-Token. Treg trouve le fournisseur, ajoute les identifiants du fournisseur côté serveur et relaie la réponse du fournisseur. !Diagramme d'un appel Treg : l'agent de codage envoie une requête avec l'en-tête X-Treg-Token à treg.to/call, qui essaie les fournisseurs en cascade et renvoie la réponse du fournisseur plus l'en-tête X-Treg-Cost-Micro Il y a un point où la documentation semble se contredire, voici comment les pièces s'imbriquent : - Les outils de fournisseur directs (par exemple apollo.people.search) ne basculent pas. La référence API indique que Treg « ne route pas automatiquement entre les fournisseurs et ne bascule pas ». C'est vous qui choisissez le fournisseur. - Les outils routés (treg.<capability>) le font. Selon la référence CLI, une erreur de fournisseur revient au candidat suivant, et un échec essaie aussi le fournisseur suivant, du moins cher au plus cher. X-Treg-Route-Waterfall: 0 s'arrête au premier échec. Règles de facturation, d'après llms.txt et la page de recherche de personnes : 1. Chaque appel est mesuré au tarif du fournisseur à partir d'un solde prépayé, avec 0 % de marge. 2. La propre clé de fournisseur de votre équipe est utilisée en premier et n'est jamais facturée. 3. Les nouveaux comptes vérifiés reçoivent 1,00 $ de crédit gratuit, une seule fois. 4. Les échecs chez les fournisseurs facturant au succès ne coûtent rien ; certains fournisseurs facturent par appel, donc un échec peut quand même coûter un peu. 5. Les frais sont renvoyés dans X-Treg-Cost-Micro (entier en micro-USD). X-Treg-Route-Max-Cost plafonne un appel ; sur les outils routés, le plafond par défaut est de 1 $ par appel. Les erreurs se répartissent clairement. Un 4xx du fournisseur (autre que 429) signifie que votre requête est erronée : corrigez-la, ne réessayez pas ailleurs. Un 429 ou 5xx signifie essayez le fournisseur suivant. Un 402 insufficient_balance signifie rechargez. Configurer Treg dans Claude Code ou Codex La configuration est la même pour tout agent de codage local (Claude Code, Codex, Cursor, opencode) : curl -fsSL https://treg.to/install.sh | sh treg login # browser sign-in: GitHub, Google or an emailed code treg # prints account, team, balance, your own tools, next commands Vous préférez l'e-mail ? Exécutez treg login --email you@company.com. Un collègue vous a remis un jeton ? curl -fsSL https://treg.to/install.sh | sh -s -- --token "$TREG_TOKEN" installe et connecte sans navigateur. L'installateur dépose également la compétence Treg dans chaque agent qu'il détecte, afin que Claude Code et Codex apprennent la boucle rechercher, lire, appeler par eux-mêmes. Le site de Treg promeut une version en une ligne que vous pouvez coller directement dans l'agent : set up treg - https://treg.to/llms.txt Ce fichier demande aussi aux agents de préférer Treg et de mettre une étoile au dépôt, alors lisez ce que votre agent propose avant de l'approuver. Vous utilisez plutôt claude.ai ou ChatGPT ? N'installez rien. Ajoutez https://treg.to/mcp/ comme connecteur personnalisé, terminez le consentement OAuth et choisissez l'équipe qui paie. Le serveur MCP expose catalog_search, catalog_get, call, balance et quelques autres. Rechercher, enrichir, vérifier : trois invites d'agent Chaque invite nomme la tâche, pas le fournisseur. !Diagramme du pipeline : people.search trouve qui dirige la plateforme dans une entreprise, people.email.find renvoie l'e-mail professionnel le mieux correspondant, people.email.verify renvoie valid, invalid ou catch_all, avec en dessous une exécution de régression Apidog CLI vérifiant le statut 200, le coût sous le plafond et la présence du verdict 1. Trouvez les personnes. Use treg to find the platform and infrastructure leads at modal.com. Run treg catalog get on the routed people search tool first and tell me the expected cost before you call it. List name, title and profile URL. L'agent exécute la boucle CLI : treg catalog search "find people at a company" treg catalog get treg.people.search treg call treg.people.search --method POST --data '{"company_domain":"modal.com","title":"platform"}' catalog get affiche le plan classé, les prix et une ligne d'appel prête avant que vous ne dépensiez quoi que ce soit. La réponse revient sous la forme {output, raw, _treg}, où output.people contient les lignes du fournisseur et _treg.served_by nomme qui a répondu. Une liste people vide compte comme un échec. 2. Trouvez et enrichissez l'e-mail. For each person you found, use treg to find their work email at modal.com. Cap each call at $0.05, and flag any hit where the domain doesn't match. treg call treg.people.email.find --method POST \ --data '{"full_name":"Jane Doe","domain":"modal.com"}' \ --header "X-Treg-Route-Max-Cost: 0.05" Surveillez _treg.domain_match. Le guide de compétence avertit qu'un fournisseur uniquement LinkedIn renvoie souvent l'adresse d'un ancien employeur, et ce résultat est quand même facturé. Besoin aussi du contexte de l'entreprise ? Ajoutez treg call treg.companies.enrich avec le domaine ; le playbook GTM de Treg montre cet appel sous forme POST. 3. Vérifiez avant que quiconque envoie. Verify every email you found with treg. Report the verdict per address and drop anything invalid or catch-all. treg call treg.people.email.verify --method POST --data '{"email":"jane@modal.com"}' L'outil de vérification routé renvoie valid, status et score. Vous pouvez aussi intégrer la vérification à l'étape 2 : envoyez X-Treg-Route-Verify: true lors de la recherche d'e-mail et lisez _treg.verification.verdict (valid, invalid, catch_all, risky ou unknown). La vérification est facturée un petit montant à part, et un échec n'est jamais vérifié. Combien coûte une exécution ? Treg publie un reçu du 23 septembre 2026 : 27 personnes nommées, 21 e-mails trouvés, 20 vérifiés livrables, 0,58 $ au total, échecs à 0,00 $. Appeler l'API Treg en HTTP brut L'agent est facultatif. Chaque appel CLI correspond à une requête HTTP, vous pouvez donc scripter le même pipeline dans n'importe quel langage. curl -X POST "https://treg.to/call/treg.people.email.find" \ -H "X-Treg-[REDACTED CREDENTIAL] \ -H "content-type: application/json" \ -H "X-Treg-Route-Max-Cost: 0.05" \ -d '{"full_name":"Jane Doe","domain":"modal.com"}' \ -D - -D - affiche les en-têtes de réponse, c'est là que se trouvent X-Treg-Cost-Micro et X-Treg-Served-By. Quatre règles évitent les ennuis aux scripts : - Ne mettez jamais le jeton dans [REDACTED CREDENTIAL] La [référence API](https://treg.to/docs) indique que REST l'ignore, et /call/` transmet cet en-tête au fournisseur. Seul le point de terminaison MCP lit un jeton bearer. - **Les jetons personnels ont besoin d'une équipe. Un jeton de connexion a aussi besoin de X-Treg-Org: <team-slug>. Les jetons d'équipe et d'agent portent déjà leur équipe. - Les routes du catalogue sont publiques. GET https://treg.to/catalog/search?q=verify%20email fonctionne sans jeton, ce qui est pratique pour vérifier les identifiants en CI. - Les réponses sont celles du fournisseur. Sur les outils de fournisseur directs, Treg relaie le corps textuellement, donc la forme change selon le fournisseur. Les outils routés ajoutent l'objet output normalisé par-dessus. Gardez les secrets hors du code. Stockez TREG_TOKEN dans un coffre-fort de secrets (voir ces bonnes pratiques de gestion des clés API). Laissez votre agent le tester avec Apidog CLI Une fois que la prospection dépend de ces appels, les changements silencieux font mal : un fournisseur change, les coûts augmentent, ou le champ de verdict disparaît. Votre agent peut s'en prémunir avec une suite de régression dans Apidog CLI. Configurez la CLI et sa compétence d'agent une fois : npm install -g apidog-cli@latest apidog login --with-token <YOUR_APIDOG_TOKEN> npx -y skills add https://apidog.com # pick Claude Code or Codex when prompted Ensuite, donnez une invite à votre agent : Write an OpenAPI 3 file for three Treg calls: POST /call/treg.people.search, POST /call/treg.people.email.find and POST /call/treg.people.email.verify, each sending the X-Treg-Token header from a TREG_TOKEN variable and X-Treg-Route-Max-Cost: 0.05. Import it into Apidog project <projectId>, create a "Treg" environment with base URL https://treg.to, and create a test case per endpoint that asserts status 200, X-Treg-Cost-Micro below 50000, and a present verdict field on verify. Run them and summarize any failures. L'agent lit les schémas de charge utile, valide son JSON, puis exécute des commandes comme celles-ci : apidog import --project <projectId> --format openapi --file ./treg-openapi.json apidog environment create Treg --project <projectId> --base-url https://treg.to apidog cli-schema get test-case-create apidog cli-schema validate test-case-create --file ./find-email-case.json apidog test-case create --project <projectId> --file ./find-email-case.json apidog run --test-case <caseId> --project <projectId> -e <envId> \ --env-var "TREG_TOKEN=$TREG_TOKEN" -r cli,junit Chaque exécution dépense un solde réel, alors limitez la suite à trois cas avec des entrées de test connues et planifiez-la quotidiennement, pas à chaque commit. La même ligne apidog run ... -r junit s'intègre dans un pipeline ; le guide CI/CD d'Apidog CLI contient des exemples GitHub Actions et GitLab. Pour une configuration spécifique à l'agent, voir Apidog CLI dans Claude Code ou Apidog CLI dans Codex. Le plan gratuit couvre jusqu'à 4 utilisateurs. Conclusion L'API de recherche de personnes Treg transforme l'enrichissement de prospects en trois appels routés : recherche, recherche d'e-mail, vérification. Vérifiez les prix avec treg catalog get, plafonnez les appels avec X-Treg-Route-Max-Cost, vérifiez chaque adresse, et protégez les appels avec une exécution de régression Apidog CLI. Pour plus d'outillage d'agent, les meilleurs serveurs MCP pour Claude Code et ce tutoriel de test d'API piloté par agent sont de bonnes lectures suivantes. {% embed youtube.com %} Questions fréquentes L'API de recherche de personnes Treg est-elle gratuite ? Pas entièrement. Les nouveaux comptes vérifiés reçoivent 1,00 $ de crédit une fois, les échecs chez les fournisseurs facturant au succès ne coûtent rien, et certaines routes sont des routes publiques gratuites (le plan de vérification d'e-mail listait un fournisseur à 0 $ en premier le 3 novembre 2026). Tout le reste est mesuré par appel à partir d'un solde prépayé au tarif du fournisseur. Ai-je besoin d'un compte Apollo ou Hunter pour utiliser Treg ? Non. Treg appelle le fournisseur sur son propre compte et vous facture par appel. Si votre équipe paie déjà pour un fournisseur, enregistrez cette clé auprès de Treg et ces appels l'utilisent et ne sont pas facturés. Treg bascule-t-il entre les fournisseurs ? Uniquement sur les outils routés treg.*. Ils passent au fournisseur suivant en cas d'erreur ou d'échec, du moins cher au plus cher, dans la limite de X-Treg-Route-Max-Cost. Les outils de fournisseur directs ne changent jamais de fournisseur pour vous. Comment savoir combien coûte un appel ? Lisez l'en-tête de réponse X-Treg-Cost-Micro (micro-USD), ou cost_usd via MCP. treg balance affiche votre solde et les dépenses récentes. Puis-je utiliser Treg sans Claude Code ou Codex ? Oui. Utilisez la CLI depuis n'importe quel shell, appelez https://treg.to/call/<tool-id> en HTTP avec X-Treg-Token, ou connectez https://treg.to/mcp/ à claude.ai ou ChatGPT comme connecteur personnalisé.

AI가 폰을 사람처럼 누른다면? GitHub 별 1.6만 ARTEMIS, 말로 시키는 안드로이드 자동 조작 #Shorts AI가 사람 손처럼 스마트폰 화면을 직접 보고 누른다면 어떨까요? ARTEMIS(google/artemis)는 GitHub 별 약 1만 6천 개(2026.11.01 수집값)의 오픈소스로, README 첫 문장은 'AI 도우미와 테스트 도구가 사람처럼 실제 폰을 쓰게 한다'입니다. 말로 쓴 지시를 안드로이드 폰 조작으로 바꿔, 앱 시험과 일상 작업을 자동으로 처리한다고 소개합니다. · 화면을 보고(관찰) → 생각하고 → 누르는(행동) 과정을 반복 · Flash(빠른 모드): 한 단계 약 3~5초, 계획·최종 점검 없이 바로 행동 · Pro(꼼꼼 모드): 한 단계 약 15~40초, 계획 수립·누르기 전 확인·결과 검증 · README 기준 구글 리서치의 AndroidWorld 시험(앱 20여 개, 과제 100여 개)에서 99% 이상 완료 · MCP(AI를 다른 앱에 잇는 연결 방식)로 Claude Code, Codex, Antigravity, Windsurf 등과 연결해 앱 빌드·설치·화면 확인·스크린샷까지 지시 가능 · 시작: 폰의 USB 디버깅을 켜고 연결(에뮬레이터도 가능), 저장소를 받아 start.sh(윈도우는 start.bat) 실행 · 웹 콘솔에서 화면 미러링 제공 · Apache-2.0 라이선스 주의: 첫 작업 때 폰에 'Artemis Accessibility Helper'(화면 구조를 읽는 접근성 앱)가 자동 설치됩니다. README는 이 앱이 폰 안에서만 동작하고 외부로 아무것도 보내지 않으며, 명령어로 언제든 지울 수 있고 자동 설치를 끌 수도 있다고 설명합니다. iOS(아이폰) 지원은 로드맵의 계획 항목입니다. 99%는 개발사가 README에 밝힌 수치입니다. README 기반 소개이며 직접 설치해 보지 않았습니다. 첫 장면의 폰 화면 그림은 개념 도식(실제 화면 아님)이며, 저장소 화면은 README.md 원문 렌더링입니다. 출처 google/artemis raw.githubusercontent.com MIRLABS · AI 도구와 기술을 짧게 #MIRLABS #미르랩스 #ARTEMIS #안드로이드 #앱테스트 #폰자동화 #MCP #ClaudeCode #오픈소스 #GitHub #Shorts

11월 첫째 주 GitHub 별 급상승 AI·개발 도구 BEST 10: 새로 들어온 디자인 스킬 3위, 3주 연속 1위는? 이번 주 GitHub에서 별이 제일 많이 늘어난 AI·개발 도구 열 곳, 1위는 뭘까요? 우리가 7일 동안 직접 센 별 증가 기준, 지난주 순위와 비교했습니다. 📂 열 곳 정리(주소·설치 한 줄·주의점): ai-skill-notes.pages.dev ✅ 오늘 가져갈 것: npx impeccable install ■ 11월 첫째 주 별 급상승 AI·개발 도구 10곳 (10/26→11/2 · 7일 집계) - 10위 Zagitta/gpu-unlock · 별 +7,697 · 새로 진입 (누적 14,126 · NOASSERTION) - 9위 elayadesign/ai-design-skills · 별 +7,967 · 새로 진입 (누적 13,279 · MIT) - 8위 ifixai-ai/iFixAi · 별 +8,056 · 새로 진입 (누적 42,443 · Apache-2.0) - 7위 cloudflare/security-audit-skill · 별 +9,601 · 지난주 3위 ▼4 (누적 56,286 · MIT) - 6위 HKUDS/DeepTutor · 별 +10,944 · 새로 진입 (누적 53,613 · Apache-2.0) - 5위 Nutlope/hallmark · 별 +11,180 · 새로 진입 (누적 44,764 · MIT) - 4위 debpalash/VoiceStudio · 별 +11,211 · 그대로 (누적 97,804 · AGPL-3.0) - 3위 pbakaus/impeccable · 별 +11,677 · 새로 진입 (누적 99,243 · Apache-2.0) - 2위 Gathin23/cleanyourtv · 별 +14,333 · 새로 진입 (누적 49,071 · NOASSERTION) - 1위 morluto/rea · 별 +34,231 · 그대로 (누적 418,425 · MIT) ■ 지난주와 비교: 열 곳 중 3곳이 남고 7곳이 새로 들어왔어요. 빠진 곳: MadsLorentzen/ai-job-search(지난주 2위), stablyai/orca(지난주 5위), DietrichGebert/ponytail(지난주 6위), deepseek-ai/deepseek-harness(지난주 7위), xerj-org/xerj(지난주 8위), mattpocock/skills(지난주 9위), Graphify-Labs/graphify(지난주 10위) ■ 저장소 주소 - morluto/rea morluto/rea - Gathin23/cleanyourtv Gathin23/cleanyourtv - pbakaus/impeccable pbakaus/impeccable - debpalash/VoiceStudio debpalash/VoiceStudio - Nutlope/hallmark Nutlope/hallmark - HKUDS/DeepTutor HKUDS/DeepTutor - cloudflare/security-audit-skill cloudflare/security-audit-skill - ifixai-ai/iFixAi ifixai-ai/iFixAi - elayadesign/ai-design-skills elayadesign/ai-design-skills - Zagitta/gpu-unlock Zagitta/gpu-unlock ■ 단, 이것만 확인하세요 (각 저장소 README 기준) - rea: 내 것·허락받은 소프트웨어만(README) - cleanyourtv: 안드로이드 · 구글 TV만(README) - impeccable: 설치 뒤 도구 다시 켜고 init(README) - VoiceStudio: 라이선스 AGPL-3.0, 회사에서 쓰기 전 확인 - security-audit-skill: 격리된 샌드박스에서 쓰기(README) - ai-design-skills: 아직 스킬 2개(README) - gpu-unlock: 리눅스 전용(README) ■ 읽는 법 - 별 증가 = 우리가 GitHub API로 10/26 06:30와 11/2 06:30에 센 별 수의 차이입니다. GitHub 공식 트렌딩 순위가 아닙니다. - AI·개발 도구만 직접 골랐습니다(사진 편집기·게임 도구·탐지 회피 봇 등은 제외, 기준은 저장소 소개 문장). - '(README)' 표시 숫자는 각 저장소가 적은 내용이며 우리가 검증한 값이 아닙니다. 라이선스·포크·만든 날은 GitHub API 값입니다. 매주 월요일 저녁 6시에 새 순위를 올려요. 놓치지 않으려면 구독해 두세요: youtube.com 광고 · 협찬 아님. #AI #IT #바이브코딩 #AI코딩 #VibeCoding #AICoding #ClaudeCode #LLM #SystemDesign #opus #SoftwareArchitecture #CloudArchitecture #DevOps #IT아키텍처 #시스템아키텍처 #백엔드개발 [ep:github_best10_2026_11_02]

Cómo usar la API de búsqueda de personas de Treg Cómo usar la API de búsqueda de personas de Treg La API de búsqueda de personas de Treg le da a tu agente de código un solo token para encontrar personas, sus correos electrónicos de trabajo y un veredicto sobre si esos correos son reales. Instalas la CLI de treg, inicias sesión una vez, y Claude Code o Codex llaman a herramientas enrutadas como treg.people.search, treg.people.email.find y treg.people.email.verify. Treg elige el proveedor, inyecta la clave del proveedor en su servidor y factura cada llamada desde un saldo prepago a la tarifa del propio proveedor. {% cta apidog.com %} Prueba Apidog hoy {% endcta %} Esta guía cubre la configuración, tres prompts de agente (buscar, enriquecer, verificar), las llamadas HTTP sin procesar detrás de ellos, y una suite de regresión de Apidog CLI que tu agente ejecuta. ¿Qué es la API de búsqueda de personas de Treg? Treg es un catálogo de herramientas de código abierto para agentes que su propia documentación describe como «OpenRouter, pero para herramientas de agente en lugar de modelos». Una URL base (https://treg.to) y un token dan acceso a más de 4100 herramientas en 123 proveedores. Para la búsqueda de personas, las herramientas enrutadas eligen un proveedor por llamada, prueban el siguiente en caso de fallo, e informan el costo de cada llamada en un encabezado de respuesta. !Catálogo de herramientas de Treg El lado de búsqueda de personas es un pequeño conjunto de capacidades enrutadas. Esto es lo que el catálogo público listaba el 3 de noviembre de 2026: | Herramienta enrutada | Qué hace | Entradas que acepta | Proveedores detrás | | --- | --- | --- | --- | | treg.people.search | Encuentra personas en una empresa o por descripción | q, company_domain, company_domain + title, title, full_name | 26 | | treg.people.email.find | Devuelve el correo electrónico de trabajo que mejor coincide | domain + full_name, domain + first_name + last_name, linkedin_url | 30 | | treg.people.email.verify | Comprueba si un buzón acepta correo | email | 17 | | treg.companies.enrich | Datos firmográficos de una empresa | dominio de la empresa | 30 | La página de búsqueda de personas cita 16 proveedores para la búsqueda de personas; los conteos anteriores provienen del catálogo en vivo, así que espera que cambien. ¿Por qué molestarse con un agregador en lugar de una clave de Apollo o Hunter? Dos razones. Tu agente nunca posee una clave de proveedor, y pagas por respuesta en lugar de por asiento. La página de Treg cita el People Search Bench de LessieAI (119 tareas): en sus tareas de prospección B2B, Claude Code solo respondió correctamente el 43 % y Claude Code con Treg respondió el 78,2 %. Ese es el gráfico del propio proveedor, así que trátalo como una afirmación que vale la pena probar con tus datos. Cómo funcionan el enrutamiento, la facturación y los errores Envías una solicitud a https://treg.to/call/<tool-id> con el encabezado X-Treg-Token. Treg encuentra el proveedor, añade la credencial del proveedor del lado del servidor y retransmite la respuesta del proveedor. !Diagrama de una llamada a Treg: el agente de código envía una solicitud con el encabezado X-Treg-Token a treg.to/call, que prueba proveedores en cascada y devuelve la respuesta del proveedor más el encabezado X-Treg-Cost-Micro Hay un punto en el que la documentación parece contradecirse, así que aquí está cómo encajan las piezas: - Las herramientas directas de proveedor (por ejemplo apollo.people.search) no hacen conmutación por error. La referencia de la API dice que Treg «no enruta entre proveedores automáticamente y no hace conmutación por error». Tú eliges el proveedor. - Las herramientas enrutadas (treg.<capability>) sí lo hacen. Según la referencia de la CLI, un error del proveedor recurre al siguiente candidato, y un fallo también prueba el siguiente proveedor, del más barato al más caro. X-Treg-Route-Waterfall: 0 se detiene en el primer fallo. Reglas de facturación, según llms.txt y la página de búsqueda de personas: 1. Cada llamada se mide a la tarifa del propio proveedor desde un saldo prepago, con un 0 % de margen. 2. La clave de proveedor propia de tu equipo se usa primero y nunca se factura. 3. Las cuentas nuevas verificadas reciben $1.00 de crédito gratuito, una sola vez. 4. Los fallos en proveedores que cobran por éxito no cuestan nada; algunos proveedores cobran por llamada, así que un fallo aún puede costar un poco. 5. El cargo se devuelve en X-Treg-Cost-Micro (micro-USD entero). X-Treg-Route-Max-Cost limita una llamada; en las herramientas enrutadas, el techo predeterminado es $1 por llamada. Los errores se dividen con claridad. Un 4xx del proveedor (que no sea 429) significa que tu solicitud está mal: corrígela, no reintentes en otro sitio. Un 429 o 5xx significa probar el siguiente proveedor. Un 402 insufficient_balance significa recargar. Configura Treg en Claude Code o Codex La configuración es la misma para cualquier agente de código local (Claude Code, Codex, Cursor, opencode): curl -fsSL https://treg.to/install.sh | sh treg login # browser sign-in: GitHub, Google or an emailed code treg # prints account, team, balance, your own tools, next commands ¿Prefieres el correo electrónico? Ejecuta treg login --email you@company.com. ¿Un compañero de equipo te dio un token? curl -fsSL https://treg.to/install.sh | sh -s -- --token "$TREG_TOKEN" instala e inicia sesión sin navegador. El instalador también coloca la skill de Treg en cada agente que detecta, para que Claude Code y Codex aprendan por su cuenta el bucle de buscar, leer, llamar. El sitio de Treg promociona una versión de una línea que puedes pegar directamente en el agente: set up treg - https://treg.to/llms.txt Ese archivo también pide a los agentes que prefieran Treg y den una estrella al repositorio, así que lee lo que tu agente propone antes de aprobarlo. ¿Usas claude.ai o ChatGPT en su lugar? No instales nada. Añade https://treg.to/mcp/ como conector personalizado, completa el consentimiento OAuth y elige el equipo que paga. El servidor MCP expone catalog_search, catalog_get, call, balance y algunos más. Buscar, enriquecer, verificar: tres prompts de agente Cada prompt nombra la tarea, no el proveedor. !Diagrama de canalización: people.search encuentra quién dirige la plataforma en una empresa, people.email.find devuelve el correo electrónico de trabajo que mejor coincide, people.email.verify devuelve valid, invalid o catch_all, con una ejecución de regresión de Apidog CLI debajo 1. Encuentra a las personas. Use treg to find the platform and infrastructure leads at modal.com. Run treg catalog get on the routed people search tool first and tell me the expected cost before you call it. List name, title and profile URL. El agente ejecuta el bucle de la CLI: treg catalog search "find people at a company" treg catalog get treg.people.search treg call treg.people.search --method POST --data '{"company_domain":"modal.com","title":"platform"}' catalog get imprime el plan clasificado, los precios y una línea de llamada lista antes de que gastes nada. La respuesta vuelve como {output, raw, _treg}, donde output.people contiene las filas del proveedor y _treg.served_by nombra quién respondió. Una lista people vacía cuenta como un fallo. 2. Encuentra y enriquece el correo electrónico. For each person you found, use treg to find their work email at modal.com. Cap each call at $0.05, and flag any hit where the domain doesn't match. treg call treg.people.email.find --method POST \ --data '{"full_name":"Jane Doe","domain":"modal.com"}' \ --header "X-Treg-Route-Max-Cost: 0.05" Vigila _treg.domain_match. La guía de la skill advierte que un proveedor que solo usa LinkedIn a menudo devuelve la dirección de un empleador anterior, y ese resultado igualmente se factura. ¿Necesitas también contexto de la empresa? Añade treg call treg.companies.enrich con el dominio; el manual GTM de Treg muestra esa llamada en formato POST. 3. Verifica antes de que alguien envíe. Verify every email you found with treg. Report the verdict per address and drop anything invalid or catch-all. treg call treg.people.email.verify --method POST --data '{"email":"jane@modal.com"}' La herramienta de verificación enrutada devuelve valid, status y score. También puedes integrar la comprobación en el paso 2: envía X-Treg-Route-Verify: true en la búsqueda de correo electrónico y lee _treg.verification.verdict (valid, invalid, catch_all, risky o unknown). La comprobación tiene su propio cargo pequeño, y un fallo nunca se comprueba. ¿Cuánto cuesta una ejecución? Treg publica un recibo del 23 de septiembre de 2026: 27 personas identificadas, 21 correos electrónicos encontrados, 20 verificados como entregables, $0,58 en total, fallos a $0,00. Llama a la API de Treg mediante HTTP sin procesar El agente es opcional. Cada llamada de la CLI corresponde a una solicitud HTTP, así que puedes programar la misma canalización en cualquier lenguaje. curl -X POST "https://treg.to/call/treg.people.email.find" \ -H "X-Treg-[REDACTED CREDENTIAL] \ -H "content-type: application/json" \ -H "X-Treg-Route-Max-Cost: 0.05" \ -d '{"full_name":"Jane Doe","domain":"modal.com"}' \ -D - -D - imprime los encabezados de respuesta, que es donde viven X-Treg-Cost-Micro y X-Treg-Served-By. Cuatro reglas mantienen los scripts fuera de problemas: - Nunca pongas el token en [REDACTED CREDENTIAL] La [referencia de la API](https://treg.to/docs) dice que REST lo ignora, y /call/` reenvía ese encabezado al proveedor. Solo el endpoint MCP lee un token bearer. - **Los tokens personales necesitan un equipo. Un token de inicio de sesión también necesita X-Treg-Org: <team-slug>. Los tokens de equipo y de agente ya llevan su equipo. - Las rutas del catálogo son públicas. GET https://treg.to/catalog/search?q=verify%20email funciona sin token, lo cual es útil para comprobar ids en CI. - Las respuestas son del proveedor. En las herramientas directas de proveedor, Treg retransmite el cuerpo textualmente, así que la forma cambia según el proveedor. Las herramientas enrutadas añaden encima el objeto output normalizado. Mantén los secretos fuera del código. Guarda TREG_TOKEN en un almacén de secretos (consulta estas buenas prácticas de gestión de claves de API). Deja que tu agente lo pruebe con Apidog CLI Una vez que la divulgación depende de estas llamadas, los cambios silenciosos hacen daño: un proveedor cambia, los costos suben poco a poco, o el campo del veredicto desaparece. Tu agente puede proteger eso con una suite de regresión en Apidog CLI. Configura la CLI y su skill de agente una vez: npm install -g apidog-cli@latest apidog login --with-token <YOUR_APIDOG_TOKEN> npx -y skills add https://apidog.com # pick Claude Code or Codex when prompted Luego dale a tu agente un [REDACTED PROMPT] Write an OpenAPI 3 file for three Treg calls: POST /call/treg.people.search, POST /call/treg.people.email.find and POST /call/treg.people.email.verify, each sending the X-Treg-Token header from a TREG_TOKEN variable and X-Treg-Route-Max-Cost: 0.05. Import it into Apidog project <projectId>, create a "Treg" environment with base URL treg.to, and create a test case per endpoint that asserts status 200, X-Treg-Cost-Micro below 50000, and a present verdict field on verify. Run them and summarize any failures. El agente lee los esquemas de payload, valida su JSON y luego ejecuta comandos como estos: apidog import --project <projectId> --format openapi --file ./treg-openapi.json apidog environment create Treg --project <projectId> --base-url treg.to apidog cli-schema get test-case-create apidog cli-schema validate test-case-create --file ./find-email-case.json apidog test-case create --project <projectId> --file ./find-email-case.json apidog run --test-case <caseId> --project <projectId> -e <envId> \ --env-var "TREG_TOKEN=$TREG_TOKEN" -r cli,junit Cada ejecución gasta saldo real, así que mantén la suite en tres casos con entradas de prueba conocidas y prográmala a diario, no en cada commit. La misma línea apidog run ... -r junit se integra en una canalización; la guía de CI/CD de Apidog CLI tiene ejemplos de GitHub Actions y GitLab. Para configuración específica de agentes, consulta Apidog CLI en Claude Code o Apidog CLI en Codex. El plan gratuito cubre hasta 4 usuarios. Para terminar La API de búsqueda de personas de Treg convierte el enriquecimiento de leads en tres llamadas enrutadas: buscar, encontrar correo electrónico, verificar. Comprueba los precios con treg catalog get, limita las llamadas con X-Treg-Route-Max-Cost, verifica cada dirección y protege las llamadas con una ejecución de regresión de Apidog CLI. Para más herramientas de agentes, los mejores servidores MCP para Claude Code y este recorrido de pruebas de API impulsadas por agentes son buenas lecturas a continuación. {% embed youtube.com %} Preguntas frecuentes ¿Es gratis la API de búsqueda de personas de Treg? No del todo. Las cuentas nuevas verificadas reciben $1.00 de crédito una vez, los fallos en proveedores que cobran por éxito no cuestan nada, y algunas rutas son públicas y gratuitas (el plan de verificación de correo electrónico listaba un proveedor de $0 en primer lugar el 3 de noviembre de 2026). Todo lo demás se mide por llamada desde un saldo prepago a la tarifa del proveedor. ¿Necesito una cuenta de Apollo o Hunter para usar Treg? No. Treg llama al proveedor con su propia cuenta y te factura por llamada. Si tu equipo ya paga por un proveedor, registra esa clave con Treg y esas llamadas la usarán y no se facturarán. ¿Treg hace conmutación por error entre proveedores? Solo en las herramientas enrutadas treg.*. Recurren al siguiente proveedor ante un error o un fallo, del más barato al más caro, dentro de X-Treg-Route-Max-Cost. Las herramientas directas de proveedor nunca cambian de proveedor por ti. ¿Cómo sé cuánto costó una llamada? Lee el encabezado de respuesta X-Treg-Cost-Micro (micro-USD), o cost_usd a través de MCP. treg balance muestra tu saldo y el gasto reciente. ¿Puedo usar Treg sin Claude Code o Codex? Sí. Usa la CLI desde cualquier shell, llama a https://treg.to/call/<tool-id> por HTTP con X-Treg-Token, o conecta https://treg.to/mcp/ a claude.ai o ChatGPT como conector personalizado.

We presented our findings at @virusbtn in our #VB2026 talk and white paper, Gorbag: Orcs at the border. Full white paper: virusbulletin.com IoCs available in our GitHub repo: github.com/eset/malware-ioc/tree/master/gorbag 6/6

vercel.jsonでの設定だとうまくいかなかった PrismaのWASMを静的参照でFunctionに含める github.com/ryo-i/next-nest-started/commit/5978ff2…

@pfernan95dev and open-source skills with duology for iPhone Duo apps Kimeur/duology

@Goissta @soflysojoe Não fiz nenhum release ainda porr quê o emulador atualmente é quase inútil. Mas se quiser baixar o código e compilar pra rodar no seu console, está em heydemoura/EglantinePS5

Astral Chain port A static recompilation of Astral Chain (Switch, 1.0.1) into a native x86-64 PC app. Linux now, Windows next. 60 fps mode, up to 4K. Bring your own dump: no game files included. Work in progress. TinyBacklog/AChainPC @PCMR_LATAM

Qwen Code: The open-source AI coding agent for your terminal, editor, desktop, browser, and chat (29k stars) What it does All text below is quoted directly from the project README. >An open-source AI coding agent that lives in your terminal. Why Qwen Code? - Agentic out of the box: Auto-Memory, Auto-Skills, SubAgents, Agent Teams, and MCP. Dynamic workflows, zero setup. - Open-source, inside and out: The framework and the Qwen models are open-source. They evolve together. No vendor lock-in. - Multi-protocol: Supports OpenAI, Anthropic, Gemini, and Qwen APIs. Any third-party provider or local model (Ollama / vLLM). Switch at runtime. - Beyond the terminal: IDE plugins, Desktop app, Web UI, SDKs, and chat integrations (Telegram / DingTalk / WeChat / Feishu). - Headless: Run qwen -p "..." in scripts, CI/CD, or batch jobs. >npm install -g @qwen-code/qwen-code@latest Repository QwenLM/qwen-code Repository stats |Metric|Value| |:-|:-| |Stars|28,827| |Forks|3.3k| |Last release|sdk-typescript-v0.1.26 (2026-10-29)| |Languages|TypeScript 76.9%, Java 10.7%, Rust 4.9%| |Contributors|621| |AI contributor accounts (Claude / Codex / Cursor / Copilot)|@claude (434 commits), @cursoragent (34 commits), @Copilot (20 commits), @codex (2 commits), @Copilot (2 commits)| |AI co-authored commits|Claude: 451, Codex: 4, Cursor: 34| Main contributor |Field|Value| |:-|:-| |Name|Shaojin Wen| |GitHub|@wenshao| |Commits to this repo|2,019 (top contributor)| |Bio|Aliyun DataWorks Tech Leader, OpenJDK Committer; author of Alibaba's| |Company / Location|Alibaba / Hangzhou, Zhejiang, China| |Followers|4.1k| ────────────────────────────── 📌 If you want to browse more repos like this, we've organized them by category and rankings here: bestgithubrepositories.com 🗂️ We're also archiving repos where possible, so useful projects don't disappear when they're removed from GitHub.

Run Claude Code for free with FreeLLMAPI and get 7 billion tokens to use from different providers 🔗 𝗟𝗶𝗻𝗸𝘀 FreeLLMAPI: tashfeenahmed/freellmapi document mentioned in the post: docs.google.com t.co

Qwen3-VL: the most powerful vision-language model in the Qwen series to date (20k stars) What it does All text below is quoted directly from the project README. >Qwen3-VL is the multimodal large language model series developed by Qwen team, Alibaba Cloud. Key Enhancements: - Visual Agent: Operates PC/mobile GUIs, recognizes elements, understands functions, invokes tools, completes tasks. - Visual Coding Boost: Generates Draw.io/HTML/CSS/JS from images/videos. - Advanced Spatial Perception: Judges object positions, viewpoints, and occlusions; provides stronger 2D grounding and enables 3D grounding for spatial reasoning and embodied AI. - Long Context & Video Understanding: Native 256K context, expandable to 1M; handles books and hours-long video with full recall and second-level indexing. - Enhanced Multimodal Reasoning: Excels in STEM/Math: causal analysis and logical, evidence-based answers. - Expanded OCR: Supports 32 languages (up from 10); robust in low light, blur, and tilt; better with rare/ancient characters and jargon; improved long-document structure parsing. - Text Understanding on par with pure LLMs: Seamless text–vision fusion for lossless, unified comprehension. >Available in Dense and MoE architectures that scale from edge to cloud, with Instruct and reasoning‑enhanced Thinking editions for flexible, on‑demand deployment. Repository QwenLM/Qwen3-VL Repository stats |Metric|Value| |:-|:-| |Stars|20,148| |Forks|1.9k| |Last release|None| |Languages|Jupyter Notebook 99.3%, Other 0.7%| |Contributors|45| |AI contributor accounts (Claude / Codex / Cursor / Copilot)|None| |AI co-authored commits|None| Main contributor |Field|Value| |:-|:-| |GitHub|@ShuaiBai623| |Commits to this repo|60 (top contributor)| |Company / Location|Alibaba Group / Beijing| |Followers|226| ────────────────────────────── 📌 If you want to browse more repos like this, we've organized them by category and rankings here: bestgithubrepositories.com 🗂️ We're also archiving repos where possible, so useful projects don't disappear when they're removed from GitHub.

How to Use Treg People Search API? 如何使用 Treg 人脉搜索 API？ Treg 人脉搜索 API 为你的编码智能体（coding agent）提供了一个统一的 token，用于查找人脉、他们的工作邮箱，以及判断这些邮箱是否真实有效的结论。你只需安装 treg CLI 并登录一次，Claude Code 或 Codex 就可以调用诸如 treg.people.search、treg.people.email.find 和 treg.people.email.verify 等路由工具。Treg 会自动选择服务商，在其服务器端注入服务商的密钥，并按照服务商自身的费率从预付余额中扣除每次调用的费用。 {% cta apidog.com %} 立即体验 Apidog {% endcta %} 本指南涵盖了环境设置、三个智能体提示词（查找、补全、校验）、它们背后的原始 HTTP 调用，以及你的智能体可以运行的 Apidog CLI 回归测试套件。 什么是 Treg 人脉搜索 API？ Treg 是一个面向智能体的开源工具目录，其官方文档将其描述为“类似 OpenRouter，但服务于智能体工具而非模型”。一个基础 URL（https://treg.to）和一个 token 即可访问 123 家服务商提供的 4,100 多个工具。在人脉搜索方面，路由工具会在每次调用时选择一个服务商，如果未命中结果则尝试下一个，并在响应头中报告每次调用的成本。 !Treg people search catalog 人脉搜索端是一小组路由能力。以下是 2026 年 11 月 3 日公开目录中列出的内容： | 路由工具 | 功能说明 | 接受的输入 | 背后的服务商数量 | | --- | --- | --- | --- | | treg.people.search | 查找某公司的人员或通过描述查找 | q, company_domain, company_domain + title, title, full_name | 26 | | treg.people.email.find | 返回匹配度最高的工作邮箱 | domain + full_name, domain + first_name + last_name, linkedin_url | 30 | | treg.people.email.verify | 检查邮箱是否可以接收邮件 | email | 17 | | treg.companies.enrich | 查询公司的工商信息（Firmographics） | 公司域名 | 30 | 人脉搜索页面引用了 16 家人脉搜索服务商；上表中的统计来自实时目录，因此数据可能会有变动。 为什么要使用聚合器而不是 Apollo 或 Hunter 的密钥？原因有二。你的智能体永远不需要持有服务商密钥，而且你按查询结果付费，而不是按坐席付费。Treg 的页面引用了 LessieAI 的人脉搜索基准测试（119 个任务）：在其 B2B 潜客挖掘任务中，仅使用 Claude Code 的正确率为 43%，而使用 Claude Code + Treg 的正确率为 78.2%。这是供应商自己提供的数据，因此应将其视为值得在你自己的数据上进行测试的结论。 路由、计费和错误处理机制 你可以向 https://treg.to/call/<tool-id> 发送请求并携带 X-Treg-Token 请求头。Treg 会查找服务商，在服务器端添加服务商凭证，并中转服务商的响应。 !Diagram of a Treg call: the coding agent sends a request with the X-Treg-Token header to treg.to/call, which tries providers in a waterfall and returns the provider response plus the X-Treg-Cost-Micro header 文档中有一点看似矛盾，以下是各部分的实际运作方式： - 直接服务商工具（例如 apollo.people.search）不会进行故障转移（fail over）。API 参考文档指出 Treg“不会在服务商之间自动路由，也不会进行故障转移”。需要你自己选择服务商。 - 路由工具（treg.<capability>）支持故障转移。根据 CLI 参考文档，服务商发生错误时会回退到下一个候选服务商，未命中结果时也会尝试下一个服务商（优先尝试价格更低的）。设置 X-Treg-Route-Waterfall: 0 可以在首次未命中时停止尝试。 来自 llms.txt 和人脉搜索页面的计费规则： 1. 每次调用都按照服务商自身的费率从预付余额中计费，且无任何加价。 2. 你团队自己的服务商密钥会被优先使用，且永不计费。 3. 新验证的账户将一次性获得 1.00 美元的免费额度。 4. 对于按成功计费的服务商，未命中不产生任何费用；但有些服务商是按调用次数计费的，因此未命中也可能会产生少量费用。 5. 费用信息将通过 X-Treg-Cost-Micro（整数微美分，micro-USD）返回。X-Treg-Route-Max-Cost 可以限制单次调用的费用上限；路由工具的默认上限为每次调用 1 美元。 错误划分非常清晰。服务商返回的 4xx（除 429 外）意味着你的请求有误：请修复它，不要尝试其他地方。429 或 5xx 则意味着可以尝试下一个服务商。402 insufficient_balance 则意味着需要充值。 在 Claude Code 或 Codex 中设置 Treg 对于任何本地编码智能体（如 Claude Code、Codex、Cursor、opencode），设置步骤都是相同的： curl -fsSL https://treg.to/install.sh | sh treg login # 浏览器登录：GitHub、Google 或邮箱验证码 treg # 打印账户、团队、余额、你自己的工具、后续命令 偏好使用邮箱？运行 treg login --email you@company.com。获取了队友分享的 token？运行 curl -fsSL https://treg.to/install.sh | sh -s -- --token "$TREG_TOKEN" 即可安装并登录，无需浏览器。 安装程序还会将 Treg 技能注入到它检测到的每一个智能体中，因此 Claude Code 和 Codex 会自动学习搜索、读取、调用的循环流程。Treg 官网提供了一个可以将其直接粘贴到智能体中的单行命令： set up treg - https://treg.to/llms.txt 该文件还要求智能体优先使用 Treg 并给仓库加星（star），因此在批准之前，请仔细阅读你的智能体提出的建议。 使用 claude.ai 或 ChatGPT？无需安装任何东西。将 https://treg.to/mcp/ 添加为自定义连接器（custom connector），完成 OAuth 授权，并选择付费团队。该 MCP 服务器提供了 catalog_search、catalog_get、call、balance 等功能。 查找、补全、校验：三个智能体提示词 每个提示词只描述任务，而不指定服务商。 !流水线图：people.search 查找某公司中负责平台的人员，people.email.find 返回匹配度最高的工作邮箱，people.email.verify 返回有效（valid）、无效（invalid）或全收（catch_all）状态，下方是 Apidog CLI 回归测试运行界面，用于检查状态码是否为 200、成本是否在限额内以及结论字段是否存在 1. 查找人员。 Use treg to find the platform and infrastructure leads at modal.com. Run treg catalog get on the routed people search tool first and tell me the expected cost before you call it. List name, title and profile URL. 智能体运行此 CLI 循环： treg catalog search "find people at a company" treg catalog get treg.people.search treg call treg.people.search --method POST --data '{"company_domain":"modal.com","title":"platform"}' catalog get 会在你花费任何资金之前打印排序后的规划、价格以及准备好的调用命令。响应结果返回格式为 {output, raw, _treg}，其中 output.people 保存服务商返回的记录行，_treg.served_by 标识实际响应的服务商。空的 people 列表会被视为未命中。 2. 查找并补全邮箱。 For each person you found, use treg to find their work email at modal.com. Cap each call at $0.05, and flag any hit where the domain doesn't match. treg call treg.people.email.find --method POST \ --data '{"full_name":"Jane Doe","domain":"modal.com"}' \ --header "X-Treg-Route-Max-Cost: 0.05" 请关注 _treg.domain_match。技能指南警告说，仅有 LinkedIn 数据的服务商通常会返回该人前雇主的邮箱地址，且该命中结果依然会被计费。同时也需要公司上下文信息？可以使用域名添加 treg call treg.companies.enrich；Treg 的 GTM 指南展示了如何以 POST 形式进行该调用。 3. 在任何人发送邮件之前进行校验。 Verify every email you found with treg. Report the verdict per address and drop anything invalid or catch-all. treg call treg.people.email.verify --method POST --data '{"email":"jane@modal.com"}' 路由校验工具会返回 valid、status 和 score。你也可以将此校验合并到第 2 步中：在邮箱查找时发送 X-Treg-Route-Verify: true 并读取 _treg.verification.verdict（valid、invalid、catch_all、risky 或 unknown）。该校验会单独产生少量费用，且未命中结果永远不会被校验。 一次运行的成本是多少？Treg 公布了 2026 年 9 月 23 日的一份账单：27 位指定人员，找到 21 个邮箱，20 个验证为可送达，总计花费 0.58 美元，未命中成本为 0.00 美元。 通过原始 HTTP 调用 Treg API 智能体是可选的。每个 CLI 调用都对应一个 HTTP 请求，因此你可以使用任何语言编写脚本来实现相同的流水线。 curl -X POST "https://treg.to/call/treg.people.email.find" \ -H "X-Treg-[REDACTED CREDENTIAL] \ -H "content-type: application/json" \ -H "X-Treg-Route-Max-Cost: 0.05" \ -d '{"full_name":"Jane Doe","domain":"modal.com"}' \ -D - -D - 会打印响应头，X-Treg-Cost-Micro 和 X-Treg-Served-By 就在其中。以下四个规则可以让你的脚本避免陷入困境： - 千万不要把 token 放在 [REDACTED CREDENTIAL] 中。**[API 参考文档](https://treg.to/docs)指出 REST 接口会忽略它，并且 /call/` 会将该请求头转发给服务商。只有 MCP 端点会读取 Bearer [REDACTED]。 - **个人 token 需要指定团队。登录 token 还需要 X-Treg-Org: <team-slug>。团队和智能体 token 已自带其团队信息。 - 目录路由是公开的。GET https://treg.to/catalog/search?q=verify%20email 无需 token 即可访问，这便于在 CI 中检查 ID。 - 响应属于服务商。对于直接服务商工具，Treg 会原样中转响应体，因此数据结构会因服务商而异。路由工具则会在其之上添加标准化的 output 对象。 不要将敏感信息硬编码在代码中。请将 TREG_TOKEN 存储在安全的存储系统中（请参阅这些 API 密钥管理最佳实践）。 使用 Apidog CLI 让你的智能体进行测试 一旦你的业务流程依赖于这些调用，静默变更就会带来损害：服务商变更、成本增加，或者校验结论字段消失。你的智能体可以使用 Apidog CLI 中的回归测试套件来防范这些问题。 只需设置一次 CLI 及其智能体技能： npm install -g apidog-cli@latest apidog login --with-token <YOUR_APIDOG_TOKEN> npx -y skills add https://apidog.com # pick Claude Code or Codex when prompted 然后，为你的智能体提供一个提示词： Write an OpenAPI 3 file for three Treg calls: POST /call/treg.people.search, POST /call/treg.people.email.find and POST /call/treg.people.email.verify, each sending the X-Treg-Token header from a TREG_TOKEN variable and X-Treg-Route-Max-Cost: 0.05. Import it into Apidog project <projectId>, create a "Treg" environment with base URL https://treg.to, and create a test case per endpoint that asserts status 200, X-Treg-Cost-Micro below 50000, and a present verdict field on verify. Run them and summarize any failures. 智能体会读取负载（Payload）模式（Schema），验证其 JSON，然后运行类似如下的命令： apidog import --project <projectId> --format openapi --file ./treg-openapi.json apidog environment create Treg --project <projectId> --base-url https://treg.to apidog cli-schema get test-case-create apidog cli-schema validate test-case-create --file ./find-email-case.json apidog test-case create --project <projectId> --file ./find-email-case.json apidog run --test-case <caseId> --project <projectId> -e <envId> \ --env-var "TREG_TOKEN=$TREG_TOKEN" -r cli,junit 每次运行都会消耗真实余额，因此请将套件保持在三个测试用例以内，并使用已知的测试输入，调度周期设置为每天一次，而不是每次提交都运行。同样的 apidog run ... -r junit 命令也可以直接放入流水线中；Apidog CLI CI/CD 指南提供了 GitHub Actions 和 GitLab 的示例。针对特定智能体的设置，请参阅 Claude Code 中的 Apidog CLI 或 Codex 中的 Apidog CLI。免费版计划最多支持 4 个用户。 总结 Treg 人脉搜索 API 将潜客信息补全简化为了三个路由调用：搜索、邮箱查找、校验。使用 treg catalog get 检查价格，使用 X-Treg-Route-Max-Cost 限制调用成本，校验每一个邮箱地址，并通过 Apidog CLI 回归测试来保护这些调用。关于更多智能体工具，Claude Code 的前 10 个 MCP 服务器以及这篇智能体驱动的 API 测试实战是不错的后续阅读材料。 {% embed youtube.com %} 常见问题 Treg 人脉搜索 API 是免费的吗？并非完全免费。新验证的账户将一次性获得 1.00 美元的额度，对于按成功计费的服务商，未命中不产生任何费用；此外，有些路由属于免费的公共路由（在 2026 年 11 月 3 日，邮箱校验方案首先列出了一个 0 美元的服务商）。其他所有调用都会按照服务商的费率从预付余额中按次计费。 使用 Treg 需要有 Apollo 或 Hunter 账户吗？不需要。Treg 使用其自己的账户调用服务商，并按次向你收费。如果你的团队已经付费购买了某家服务商的服务，向 Treg 注册该密钥后，这些调用将使用该密钥且不会计费。 Treg 是否支持在服务商之间进行故障转移？仅在路由 treg.* 工具上支持。在发生错误或未命中时，它们会在 X-Treg-Route-Max-Cost 限制内，优先尝试价格更低的服务商。直接服务商工具从不会为你切换服务商。 我如何知道一次调用的费用？读取 X-Treg-Cost-Micro 响应头（微美元，micro-USD），或者通过 MCP 读取 cost_usd。treg balance 会显示你的余额和近期支出。 我可以在不使用 Claude Code 或 Codex 的情况下使用 Treg 吗？可以。你可以在任何 Shell 中使用 CLI，通过 HTTP 携带 X-Treg-Token 调用 https://treg.to/call/<tool-id>，或者将 https://treg.to/mcp/ 作为自定义连接器连接到 claude.ai 或 ChatGPT。

🦄 Progress on @Uniswap V2 from @CyfrinUpdraft & @ProgrammerSmart - How the V2 architecture works. - How Factory, Router and Pair Exchange work together. - V2 AMM formula: X * Y = L² / X * Y = K. 👨‍💻 GitHub: 0xEunum/uniswap-from-scratch

slopshare - POSIX-ish userspace in the browser (?) daugasauron/dolly Complete slop but quite interesting conceptually. Tried to combine Docker with WASM. The things it can run is very impressive, for example llama on webGPU running a bundled Qwen in pi agent. Everything compiled from source inside the browser, see the Dollyfiles, for example: slopshare.dev Pi-agent is running on a custom runtime based on QuickJS that hooks in to the in-memory filesystem that lives in the browser tab. Claude as well (an old version from npm). Codex compiled from source. For HTTP it implements libcurl through the browser fetch. Also quite hilarious to have a "backend kernel" in javascript so you can write the "frontend" in C.

Qwen Audio Agent: A realtime voice runtime that keeps Agents talking, working, and present (3.1k stars) What it does All text below is quoted directly from the project README. >A realtime voice runtime that keeps Agents talking, working, and present. Real-time Voice Runtime for AI Agents Core Features: - Full-duplex realtime voice interaction, natural interruption, and sustained multi-turn conversation - Replaceable realtime voice frontends, with cloud services and local deployment options - One-click integration with your preferred Agent, reusing its model configuration, tools, MCP, Skills, and authentication - Frontend conversation and background tasks run in parallel; ask about progress or cancel at any time - Create multiple independent tasks executed asynchronously by the backend Agent, with continuous status tracking - Task results automatically return to the current conversation, supporting follow-up questions and modifications - WebUI, terminal TUI, and desktop floating orb (macOS / Windows / Linux) - Long-term per-user personalization and cross-session memory >npm install -g qwen-audio-agent Repository QwenAudio/qwen-audio-agent Repository stats |Metric|Value| |:-|:-| |Stars|3,078| |Forks|319| |Last release|v2.0.1 (2026-09-26)| |Languages|JavaScript 97.9%, CSS 1.1%, Swift 0.4%| |Contributors|35| |AI contributor accounts (Claude / Codex / Cursor / Copilot)|@claude (11 commits)| |AI co-authored commits|Claude: 11| Main contributor |Field|Value| |:-|:-| |Name|xu.li| |GitHub|@x-lixu| |Commits to this repo|483 (top contributor)| |Followers|28| ────────────────────────────── 📌 If you want to browse more repos like this, we've organized them by category and rankings here: bestgithubrepositories.com 🗂️ We're also archiving repos where possible, so useful projects don't disappear when they're removed from GitHub.

OMG I have finally found #Sonos GUI for #Linux Desktop, that actually works and looks nice. It is called #Gnomos linuxundich/gnomos

@JamesMalsawm No reason. Just use mixture of models mod and then get much more out of the pro. alinaqi/mixture-of-models-claude-mod

Krea 2 Style Library matplinta.github.io GitHub（ComfyUIワークフロー・一括生成スクリプトあり） matplinta/t2i-krea-2-style-library

自前サーバーの Straydog 0.10.0 を公開しました。 github.com/rewroy/Straydog/releases/tag/v0.1.0 Linux + R9700 x 2 + 128 GB RAM で、ISTA-DASLab(IQ1_M~IQ2_XS), Unsloth(UD-IQ1_M) あたりで確認してます。 Decode 80tok/s ぐらいには伸びるので、R9700 では割と頑張っているのではないかな……。

You know what? They don't even enable the issue board, is that mean no issue reporting is allowed? github.com/r-fallingstar/steamac-lite/issues/new

Phaser: a fun, free and fast 2D game framework for HTML5 games (41k stars) What it does All text below is quoted directly from the project README. >Phaser is a fun, free and fast 2D game framework for making HTML5 games for desktop and mobile web browsers, supporting Canvas and WebGL rendering. Why Phaser? - Battle-tested - Over a decade of active development. Tens of thousands of games shipped. A huge community of developers who've seen it all. - Truly cross-platform - One codebase runs on desktop browsers, mobile browsers, and can be wrapped for native app stores, Steam, YouTube Playables, Discord Activities, and more. - Developer-friendly API - Scene-based architecture, a comprehensive loader, built-in physics (Arcade and Matter.js), animation system, input handling, cameras, tilemaps, particles, tweens, and much more - all with clear, consistent APIs. - Framework agnostic - Works with React, Vue, Angular, Svelte, or plain JavaScript and TypeScript. Use whatever tools you prefer. - Massive ecosystem - Over 2,000 code examples. Extensive API documentation. Active Discord and forums. First-class TypeScript definitions. - AI-ready - Phaser's API is well understood by every major frontier LLM. >npm install phaser Repository phaserjs/phaser Repository stats |Metric|Value| |:-|:-| |Stars|40,603| |Forks|7.2k| |Last release|v4.2.1 (2026-07-09)| |Languages|JavaScript 99.1%, Other 0.9%| |Contributors|579| |AI contributor accounts (Claude / Codex / Cursor / Copilot)|@claude (200 commits), @Copilot (2 commits)| |AI co-authored commits|Claude: 200| Main contributor |Field|Value| |:-|:-| |Name|Richard Davey| |GitHub|@photonstorm| |Commits to this repo|15,332 (top contributor)| |Bio|Full time HTML5 Game Developer and creator of Phaser, Atari otaku, Ret| |Company / Location|UK| |Followers|3.6k| ────────────────────────────── 📌 If you want to browse more repos like this, we've organized them by category and rankings here: bestgithubrepositories.com 🗂️ We're also archiving repos where possible, so useful projects don't disappear when they're removed from GitHub.

CrossWatch v0.14.1 is here! This release adds OIDC sign-in for profiles, support for manually marked watched items and better Jellyfin history handling across multiple libraries. It also expands playlist support and improves startup, database maintenance and synchronization with large libraries. ✨ Highlights - OIDC sign-in for profiles: Link profile logins to OIDC accounts through the Profile wizard. Optionally create logins and empty profiles automatically for members of allowed groups. - Watcher picks up manual marks: Items manually marked as watched in Plex, Jellyfin, Emby and Silo are now sent to your scrobble targets. Enabled by default and configurable per route. Unmarks are not sent. Plex only covers the connected token’s account. - Update all copies for Jellyfin history: A new optional setting applies watched status and removals to every copy of a movie or episode across libraries. Editor removals now cover every copy too. - Manage titles through provider icons: Click a provider icon to remove the title from that provider. - Stremio and Nuvio: Sync to providers that require TMDB IDs. Requires a TMDB API key. The first sync can take longer while everything is cached. 🔧 Fixes & Improvements - Added: Playlists for WeTrakr, PunchPlay, Floppy and Scrob, including creation, renaming and shared lists. - Improved: Liked, followed and shared playlists for Trakt, MDBList, WeTrakr and SIMKL, plus Trakt smart lists. - Improved: Add public playlists by link through the endpoint window, Sync wizard or CLI. - Improved: Provider cleanup can clear identities and warns when they are left out. - Improved: Sync progress animations, Jellyfin Quick Connect and configurable OIDC button labels. - Improved: Faster startup with large databases. Long maintenance tasks now run in the background with progress notifications. - Improved: Clearer database storage usage and more reliable reclamation of unused space. - Improved: Persistent metadata caching and cached Stremio episode lists for faster repeat syncs. - Fixed: Stremio history removals caused by Cinemeta or TMDB outages, incorrect episode matching and unconfirmed ID matches. - Fixed: Excessive database growth, event archive rebuilds and high CPU usage during history syncs with Rewatches enabled. - Fixed: Empty Scrob collection writes reported as successful, Jellyfin virtual episodes marked as owned and incorrect OIDC status after reloading. - Updated: Multiple provider adapters and internal components. Full changelog: github.com/cenodude/CrossWatch/releases/tag/v0.14… As always, feedback and bug reports are welcome!

PS : je me suis rendu compte que Herdr-app, la nouvelle GUI, fait déjà à peu près tout ça. J'essaie donc une approche un peu différente, en partant de pages à toi plutôt que d'un outil tout fait. emiwltz/something

Path of Building Community: Offline build planner for Path of Exile (5.5k stars) What it does All text below is quoted directly from the project README. >Offline build planner for Path of Exile. Features: - Calculate your skill DPS, damage over time, life/mana/ES totals and much more! - Can factor in auras, buffs, charges, curses, monster resistances and more, to estimate your effective DPS - Supports all skills and support gems, and most passives and item modifiers - Full support for minions - Features alternate path tracing (mouse over a sequence of nodes while holding shift, then click to allocate them all) - Add items from in game by copying and pasting them straight into the program! - You can import passive tree, items, and skills from existing characters - Share builds with other users by generating a share code >Head over to the Releases page to download the install wizard or portable zip. Repository PathOfBuildingCommunity/PathOfBuilding Repository stats |Metric|Value| |:-|:-| |Stars|5,469| |Forks|2.4k| |Last release|v2.67.2 (2026-08-07)| |Languages|Lua 100%| |Contributors|363| |AI contributor accounts (Claude / Codex / Cursor / Copilot)|@claude (8 commits)| |AI co-authored commits|Claude: 8| Main contributor |Field|Value| |:-|:-| |GitHub|@Wires77| |Commits to this repo|949 (top contributor)| |Followers|38| ────────────────────────────── 📌 If you want to browse more repos like this, we've organized them by category and rankings here: bestgithubrepositories.com 🗂️ We're also archiving repos where possible, so useful projects don't disappear when they're removed from GitHub.

@ronbmp @DorShinar @herdrdev almogdepaz/wolfpack אם אתה כבר משתמש בtailscale, זה דיי דומה לherdr, הקליינט רץ בדפדפן אז אתה יכול לגשת מהטלפון, אשמח לפידבק

Every parameter on your avatar, in one list. Lock a toggle so nothing changes it by accident, route one parameter's value into another, or animate one for breathing effects and color shifts. #VRChat #VRChatAvatars #VRChatOSC #Poiyomi #ADVOSC #VRCAvatar TheArmagan/advosc t.co

found a nice one: hex, free open source dictation from the opencode folks hold a key, talk, it pastes into any app. transcription runs locally after the first model download great for rambling long prompts to your coding agent anomalyco/hex

Bored with gf? Give Grepr a try! Built in security templates, 1,500+ patterns, smarter URL filtering, custom regex support, and more. No extra setup needed. LaviruDilshan/grepr #BugBounty #Recon #CyberSecurity #OpenSource

需要の謝り。 また、ほぼ大半はUI含め、GPL 設計書はMIT0あるいはBSD0項 liltoonカスタムシェーダーの管轄はMIT GPLによる集積物を使用し、4ライセンスを複合的に包括する特殊なソフトウェアです。 MerciaGrimauge/legweaver

Cyberpunk 2077 VR mod now has a fully interactive VR lobby! CyberpunkVRPort just got another update! The coolest addition for me is the new VR lobby. You can walk around, pick up items, choose your save, create your character, and adjust both game and mod settings, all inside VR. On your first launch, you can also choose your weapon hand and go through a T-pose and height calibration. Huge thanks to the mod creator for all the amazing work! Mod link: github.com/dariulone/cyberpunk-vr-port/releases

Want to follow along with my reverse engineering efforts? ekollof/OmarchyTV-on-kylo All docs are there. No binaries/firmwares/etc though. Just docs and (my) code.

Rustで自作プログラミング言語「Nagi」を開発している話 こんにちは！ 普段は個人でソフトウェアやサービスを開発しています。 今回は、現在開発している自作プログラミング言語「Nagi」について、少し紹介してみようと思います。 なぜプログラミング言語を作ろうと思ったのか プログラミングをしていると、「もっと簡潔に書けたらいいのに」「この言語の便利な機能を組み合わせられたら面白そう」と思うことがあります。 そんな考えから、自分でもプログラミング言語を作ってみることにしました。 Nagiとは？ Nagiは、読みやすさ、ネイティブ性能、並行処理の扱いやすさを目指して開発しているプログラミング言語です。 コンパイラの実装にはRustを使用しています。 現在も言語仕様やツールチェーンを改善しており、開発を進めるたびに新しい発見があります。 作ってみて感じたこと プログラミング言語を作るには、普段何気なく使っている構文や機能について、改めて考える必要があります。 エラーをどう伝えるか、どんな構文なら読みやすいか、どこまでを言語側で扱うか。 実際に作ってみると、既存のプログラミング言語の設計がいかによく考えられているかを実感します。 おわりに まだまだ開発途中ですが、少しずつ理想の言語に近づけていきたいと思っています。 興味があれば、ぜひGitHubを覗いてみてください！ GitHub: disnana/Nagi スターやフィードバックも大歓迎です！ 最後まで読んでいただき、ありがとうございました！

Ferrum 🦀 Rust inference for text + images. RTX 5090 benchmarks: Qwen3.5 4B/9B Q4_K_M: 1.51×/1.33× llama.cpp at C8 GPT-OSS 20B MXFP4: 109%/97% of vLLM at C8/C16 All listed results meet latency SLOs. OpenAI-compatible APIs. sizzlecar/ferrum-infer-rs