trycua

I built an open-source platform where AI agents work like governed employees, and now they can use desktop apps Most "AI agent" demos stop at the moment it gets interesting for a company: who's allowed to let the agent pay an invoice, send an email or change a record, and how do you know what it did? That's the problem I've been working on with Agent Hangar, an open-source (Apache-2.0), self-hosted platform where agents work as digital employees. It covers the whole range: from a simple chat agent, to agents built in other frameworks, all the way to a Digital employee with a harness (Claude Code, Codex) that works in your IDE with its own role, budget and authority. GitHub: valteresj2/agent-hangar {% github valteresj2/agent-hangar %} The idea: an agent is an employee, not a script Each Digital employee has: - a role and a team, and a person it reports to; - a monthly budget in tokens and money; - a list of allowed tools (connectors, MCP servers, platform tools); - an authority level: anything above it waits for a person's approval. Every task has a timeline with each step, the decisions taken and who approved them. New: governed computer use A lot of real work still lives in desktop applications with no API: a legacy ERP, a spreadsheet, an installed system. In the latest release, a Digital employee can operate them the way a person does, by looking at the screen. !The agent clicked "Confirmar baixa" in a legacy ERP; the click had a manager's approval How it stays safe: - It runs in an isolated Linux desktop (built on Cua) with no internet access, and the agent can open only the apps allowed for it. - Each action returns the window's screenshot plus its accessible elements, numbered. Apps with no accessibility info are operated by position on the screenshot. - Clicks that save, post, send or pay go through the authority check. The approval request carries the screenshot, and the click happens exactly where the person approved. - Passwords stay in the vault. The agent types secret://name; the model and the timeline never see the value. !The ERP's status bar shows the payment recorded I tested it end to end with a real model (DeepSeek through OpenRouter) on 8 cases: logging in with a vault password, a date typed in another format, a nonexistent invoice, a confirmation dialog, menus with scrolling, saving a spreadsheet with "Save as", reading a CSV, and two payments in one task without duplicates. 8 of 8 passed. The test also caught a real bug (tool events showing twice on the timeline), fixed in 0.58.4. There's an 80-second video of the run in the README. Turning it on docker compose --profile desktop up -d desktop Then list the allowed apps in the employee's settings. Two sample apps come with it (Gnumeric and a Tk "legacy ERP"); you add your own in the desktop's Dockerfile. What else is in there - Any LLM per agent: OpenAI, Anthropic, Gemini, OpenRouter, a corporate gateway, Ollama, any OpenAI-compatible endpoint - Agent memory as a graph on PostgreSQL with Apache AGE - Warnings before a model an agent uses is deprecated - An MCP endpoint, SSO/SCIM, Helm and Terraform installs, a desktop app Feedback welcome I'd love to hear what you'd need to trust an agent with real work, and where the safety boundaries for computer use should be. Issues and PRs are open on GitHub: valteresj2/agent-hangar

Why We Run Our AI Agent on Your Mac Instead of a Cloud Computer A lot of personal AI agents now come with their own computer: a virtual machine in the cloud with a browser, a file system and a model driving it. It's a clean design. The agent gets a sandbox it fully controls, and nothing it does can touch your machine. The catch shows up on the first real task. Your files are on your laptop. Your logins live in your browser. Your mail, calendar and the odd desktop app your team uses are all local. So you upload documents, sign in to your accounts again on a remote machine, and download whatever it makes. We build Nico, a personal agent for the Mac, and we went the other way: the agent works on the Mac you're already using. This post covers the design choices that fell out of that, and a few things we learned building it. Two Agents Behind One Character Nico shows up as a small character on your desktop. You hold the right ⌘ key, say what you need, and let go. Behind that there are two agents with very different jobs. The front agent talks with you. It has to answer fast, follow speech that changes its mind halfway ("Thursday, no, Friday"), and stay available while work is running. It doesn't do long tasks itself. It works out what you mean, hands a task off, and decides when to tell you how it went. The Worker does the work. It plans, reads files, works in your apps and browser, makes what you asked for, and checks its own results. Tasks can take minutes, get queued, pause, and resume after the app restarts. !Two small Nico figures under the line “Not one agent, two agents”: one labeled “One talks with you,” the other digging through a box of documents, labeled “One does the work” Splitting them keeps voice latency independent of task length. You can ask for a report, then ask something unrelated while the report is still being written, and the front agent answers right away. Built on pi, With Our Own Tools The Worker runs on pi, an open-source agent harness. pi owns the loop that sends requests to the model, gets back tool calls and returns results. It also connects to model providers and subscription logins, saves sessions, and retries on errors. Nico owns everything around that loop: every tool the Worker can call (including replacements for pi's built-in ones), the check before each step, confirmation cards, a log of outbound actions, the system prompt, and what happens when the context fills up. That last one is a good example of why we split things this way. When the context is full, pi's default is to have the model write a summary that keeps the first 2,000 characters of each tool result. We ran a comparison: the Worker reads three long pages, each ending with a reference number, the context fills up, and then we ask for the three numbers. - With the default summary, none of the three models we tried got a single number right. Two of them also recorded emails they had already answered as still unanswered. - With our own compaction summary, all three got every number right. Our summary also attaches the outbound actions already taken, marked as done and not to be repeated. An agent that forgets it already sent an email will cheerfully send it again. Every Step Gets a Risk Level Running on your real Mac means the agent can reach real things, so every step is classified before it runs: do it directly, send it to review, ask the user, or refuse. - Review checks the step against what the task was authorized to do. Typing into a page or tidying files in a folder you approved goes straight through. Other steps go to an automatic review, and if review says no, the reason goes back to the Worker instead of stopping the task. - Asking the user is reserved for things that shouldn't be automated. Sending an email from Mail shows a card with the account, recipients, subject, full text and attachments, and it waits for a click on Send. Payments, account authorization and credentials always go back to you. - Steps with an unknown outcome (say the app quit mid-send) are checked read-only first, for example by looking in the Sent folder, before anything is retried. Some actions get a floor regardless of what the review thinks: sending, submitting, permanent deletion, sharing and privacy changes are always treated as high risk. !An expense report filled in from a hotel receipt in Chrome, with Nico asking “All filled in. Should I submit it?” and the options Submit, Not yet and Other Local Means Your Logins Just Work Because the Worker runs on your Mac: - Files are read where they are, and results are saved locally. - Websites see your usual network and your existing sessions. The Worker uses its own browser on your Mac, or your Chrome through the Nico extension. - Mail, Calendar, Reminders, Notes, Finder, Preview and other apps are within reach. Desktop apps are driven through cua-driver, and Nico decides which tools and system permissions it gets. - Memory is stored on your Mac, and you can see, change or delete it. Bring Your Own Model The Worker always runs on your own model account: ChatGPT, Claude, Gemini, GitHub Copilot, DeepSeek, Qwen, Kimi, Mistral, OpenRouter, or any OpenAI- or Anthropic-compatible endpoint. Credentials go in the macOS Keychain, and requests go straight from your Mac to the provider you picked. Nico's servers never see what the Worker reads or writes. If you've already got a coding plan or API key, there's no second AI subscription just to run tasks. Try It Nico is free for Apple silicon Macs on macOS 15.4 or later. The Free plan runs tasks on your own model and replies in text. Plus ($15/month) adds voice replies, and new accounts get a 7-day Plus trial with no card required. Get Nico at nico.app If you're building agents yourself, we'd love to hear how you handle compaction and outbound actions. Drop a comment.

2,577 Stars 2 Din Mein: AI Coding Ki Agli Chhalaang OpenAI Se Nahi? 5 GitHub Launches #Shorts Top 5 trending GitHub repos, Hindi mein - har number aaj check kiya gaya hai. Is week ke top trending repos, 75k+ stars combined: 1. Asymptote-Labs/agent-beacon (3k stars) - The cross-harness, self-improving memory layer for AI agents. Join our community: discord.com 2. trycua/cua (31k stars) - Scale computer-use 2.0 with open-source drivers, cross-OS sandboxes, and benchmarks for training, evaluation, and data generation. 3. Fei2411/DeepSeek-Bot (3k stars) - A team of long-lived bots for DeepSeek Harness, with main bots, bot-to-bot messages and group chats. 4. noahdunnagan/fsearch (2k stars) - Whole-disk file search for macOS: fuzzy names, typo tolerance, indexed content grep. ~1 ms over 8M files. 5. utmapp/UTM (36k stars) - Virtual machines for iOS and macOS Repo links: Asymptote-Labs/agent-beacon trycua/cua Fei2411/DeepSeek-Bot noahdunnagan/fsearch utmapp/UTM Comment karo "VIEWS" - saare links DM mein bhej dunga. Kisi dev dost ko bhejo jise ye chahiye. Save karlo, baad mein kaam aayega. Naman AI - AI aur dev tools, Hindi mein. Har din naya repo, har number verified. #Shorts #GitHub #AI #OpenSource #TechHindi [github trending repos, best github repositories 2026, top 5 github repos, open source projects for beginners, ai coding agents, github stars] [coding in hindi, tech shorts hindi, programming for beginners, developer tools 2026, ai tools for developers, learn coding free] [bn] ২ দিনে ২,৫৭৭ স্টার: এআই কোডিংয়ের পরবর্তী ধাপ ওপেনএআই থেকে নয়? ৫টি গিটহাব লঞ্চ #Shorts হিন্দিতে শীর্ষ ৫টি ট্রেন্ডিং গিটহাব রিপো - আজ প্রতিটি নম্বর চেক করা হয়েছে। এই সপ্তাহের শীর্ষ ট্রেন্ডিং রিপো, সম্মিলিতভাবে ৭৫ হাজারেরও বেশি স্টার: ১. Asymptote-Labs/agent-beacon (৩ হাজার স্টার) - এআই এজেন্টদের জন্য ক্রস-হারনেস, স্ব-উন্নয়নশীল মেমরি লেয়ার। আমাদের কমিউনিটিতে যোগ দিন: discord.com ২. trycua/cua (৩১ হাজার স্টার) - ওপেন-সোর্স ড্রাইভার, ক্রস-ওএস স্যান্ডবক্স এবং প্রশিক্ষণ, মূল্যায়ন ও ডেটা তৈরির জন্য বেঞ্চমার্কের মাধ্যমে কম্পিউটার-ব্যবহার ২.০-কে স্কেল করুন। ৩. Fei2411/DeepSeek-Bot (৩ হাজার স্টার) - DeepSeek Harness-এর জন্য দীর্ঘস্থায়ী বটদের একটি দল, যেখানে রয়েছে মূল বট, বট-টু-বট মেসেজ এবং গ্রুপ চ্যাট। 4. noahdunnagan/fsearch (2k stars)- macOS-এর জন্য সম্পূর্ণ-ডিস্ক ফাইল অনুসন্ধান: অস্পষ্ট নাম, টাইপো সহনশীলতা, সূচীকৃত বিষয়বস্তু grep। 8M ফাইলের উপরে ~1 ms 5. utmapp/UTM (36k তারা) - iOS এবং macOS এর জন্য ভার্চুয়াল মেশিন রেপো লিঙ্ক: Asymptote-Labs/agent-beacon trycua/cua Fei2411/DeepSeek-Bot noahdunnagan/fsearch utmapp/UTM মন্তব্য করুন "দর্শন" - সারা লিঙ্ক ডিএম মে ভেজ দুঙ্গা। কিসি দেব দোস্ত কো ভেজো জিসে ইয়ে চাহিয়ে। করলো বাঁচাও, বাদ মে কাম আয়েগা। নমন এআই - এআই অর দেব টুল, হিন্দি মে। প্রতিদিন নতুন রিপো, প্রতিটি নম্বর ভেরিফাইড। #শর্টস #গিটহাব #এআই #ওপেনসোর্স #টেকহিন্দি [গিটহাব ট্রেন্ডিং রিপো, সেরা গিটহাব রিপোজিটরি ২০২৬, সেরা ৫টি গিটহাব রিপো, নতুনদের জন্য ওপেন সোর্স প্রজেক্ট, এআই কোডিং এজেন্ট, গিটহাব স্টারস] [হিন্দিতে কোডিং, হিন্দি টেক শর্টস, নতুনদের জন্য প্রোগ্রামিং, ডেভেলপার টুলস ২০২৬, ডেভেলপারদের জন্য এআই টুলস, বিনামূল্যে কোডিং শিখুন]

Cua: Open-Source Toolkit That Gives AI Agents Their Own Computer #Shorts Cua gives AI agents real desktops they can use: click, type and work inside apps across macOS, Windows and Linux, switching between code, APIs and the GUI in one task. It includes Cua Driver, Cua Spaces, Lume local VMs and Cua Bench, connects via MCP, SDKs or CLI, and has passed 30,000 GitHub stars. Source: x.com Repo: trycua/cua Not financial advice. #AI #AIagents #OpenSource #ComputerUse #Automation

4/ To try it, install the plugin and set the backend to cua: hermes plugins install trycua/hermes-cua-spaces hermes config set terminal.backend cua Each task starts in a clean Space, so commands never touch your laptop, and Hermes can run them without asking you to approve each one.

@migdoesit no need to reverse engineer it, the code's on GitHub and there's a free plan: trycua/cua if you build something on it, show us!

5/ Fork the full bots app for Mac and iPhone: github.com/trycua/cua/tree/main/samples/cua-bots-… Docs: cua.ai What's the first job you'd give a bot running on your own computer? t.co

7/ Spaces is built on Cua Driver, the layer that lets agents operate real desktop apps. Claude Code, Codex and other agents connect through the Cua MCP server. The Spaces app is source-available under FSL-1.1-MIT: github.com/trycua/cua/tree/main/apps/cua-spaces

5 Insanely Useful GitHub Repos Blowing Up Right Now (Oct 2026) Five free GitHub repos are blowing up right now, and every one is insanely useful: an AI with its own computer, a free team of analyst agents from Anthropic, and one sentence that turns into a real part you can 3D print. 🎁 FREE: The Rising Five (all five, how to install each one, and what to watch out for). Comment RISING and my bot sends it to you. #5 Cua (+7,084★ this month): gives AI agents a Mac or Linux desktop of their own, on your Mac → trycua/cua #4 Claude for Financial Services by Anthropic (+4,838★ this month): pitch, earnings and model-builder agents for Claude → anthropics/financial-services #3 Anatomy (6.5K★ in its first 4 days): a Claude Code skill that explains any idea by building an interactive working machine → wheresryan22/anatomy #2 e2e (+5,582★ this week): app tests written as plain-English sentences, replayed with zero AI calls → tester-army/e2e #1 text-to-cad (+2,441★ this week): your AI designs real parts and exports STL / 3MF / STEP → earthtojake/text-to-cad We ran four of them ourselves (Oct 12, 2026): text-to-cad, e2e, Anatomy and the DCF agent from Claude for Financial Services. The Cua footage is Cua's own official demo video. The DCF is a demo, not financial advice. ⏱ Chapters 0:00 5 repos blowing up right now 0:22 #5 Cua: your AI gets its own computer 0:58 #4 Claude for Financial Services: free analyst agents 1:49 #3 Anatomy: ideas become working machines 2:38 #2 e2e: tests in plain English 3:30 #1 text-to-cad: one sentence → a real part 4:17 Get all five 📚 New to Claude Code? Beginner guides: • The Complete Claude Code Guide: hyperautomationlabs.gumroad.com • The OpenAI Codex Guide: hyperautomationlabs.gumroad.com • Claude Cowork Sales Guide: hyperautomationlabs.gumroad.com • Claude Certified Architect Prep Kit: hyperautomationlabs.gumroad.com ❤️ Support the channel: hit Join on YouTube, or become a Supporter on Facebook. 📸 Instagram & Facebook: Hyperautomation Labs (bite-size AI) Independent roundup, not affiliated with any project. Star counts change fast.

[오픈소스] 깃허브를 뒤흔든 미친 오픈소스 TOP 20 00:00 깃허브에서 최근 화제가 된 오픈소스 TOP 20 프로젝트 소개 00:39 20개 프로젝트를 통해 살펴보는 주요 메가 트렌드 개요 01:23 AI를 팀원으로 고용하여 업무를 협업하는 에이전트 기술 04:10 에이전트의 단기 기억 한계를 극복하는 롱텀 메모리 솔루션 05:50 데스크톱 및 인터페이스를 직접 제어하는 Computer Use 에이전트 07:34 로컬 PC 자원을 활용한 AI 비디오 생성 및 공간 데이터 시각화 08:53 개발 환경 최적화 및 경량화를 통한 코어 개발 툴의 혁신 11:13 AI 결과물 검증 프로세스 및 개인정보/권한 보안 강화 12:43 순수한 재미와 호기심에서 출발한 오픈소스 가상화 및 커스텀 혁신 14:01 자율 에이전트 시대의 도래와 인간 역할에 대한 성찰 및 마무리 The RAD Debugger Project : EpicGames/raddebugger OpenRig : mvschwarz/openrig Agent Reach : Panniantong/Agent-Reach God's Eye View : bilawalsidhu/gods-eye-view Hindsight : vectorize-io/hindsight knowledge-work-plugins : anthropics/knowledge-work-plugins cua : trycua/cua RemoveMacAI : omlahore/RemoveMacAI bloodborne_pcbloodborne_pc : deadinside28/bloodborne_pc FreeVideo : FlashML-org/FreeVideo leviathan : elstongun/leviathan Agent Memory Repo : AgentMemoryRepo/agent-memory livenerf : ninjahawk/livenerf mygo : egoist/mygo OpenDots : CopilotKit/OpenDots clearai-dsh : Clearailhc/clearaidsh omnirush-gui : omnirush-ai/omnirush-gui hairline : lucasmarkes/hairline bops : OrgoAI/bops gdp-ts : rauchg/gdp-ts [en-US] [Open Source] The Top 20 Insane Open Source Projects That Shook GitHub 00:00 Introduction to the Top 20 Open Source Projects Recently Trending on GitHub 00:39 Overview of Major Mega Trends Explored Through 20 Projects 01:23 Agent Technology for Collaborating on Work by Hiring AI as Team Members 04:10 Long-Term Memory Solutions Overcoming the Short-Term Memory Limitations of Agents 05:50 Computer Use Agents Directly Controlling Desktops and Interfaces 07:34 AI Video Generation and Spatial Data Visualization Utilizing Local PC Resources 08:53 Innovation in Core Development Tools Through Development Environment Optimization and Lightweighting 11:13 AI Output Verification Process and Enhanced Security for Personal Information and Permissions 12:43 Open Source Virtualization and Custom Innovations Starting from Pure Fun and Curiosity 14:01 The Arrival of the Autonomous Agent Era, Reflections on the Role of Humans, and a Wrap-up The RAD Debugger Project : EpicGames/raddebugger OpenRig : mvschwarz/openrig Agent Reach : Panniantong/Agent-Reach God's Eye View: bilawalsidhu/gods-eye-view Hindsight: vectorize-io/hindsight knowledge-work-plugins: anthropics/knowledge-work-plugins cua: trycua/cua RemoveMacAI: omlahore/RemoveMacAI bloodborne_pcbloodborne_pc: deadinside28/bloodborne_pc FreeVideo: FlashML-org/FreeVideo leviathan: elstongun/leviathan Agent Memory Repo: AgentMemoryRepo/agent-memory livenerf: ninjahawk/livenerf mygo: egoist/mygo OpenDots: CopilotKit/OpenDots clearai-dsh: Clearailhc/clearaidsh omnirush-gui: omnirush-ai/omnirush-gui hairline: lucasmarkes/hairline bops: OrgoAI/bops gdp-ts: rauchg/gdp-ts [id] [Open Source] 20 Open Source Gila yang Mengguncang GitHub 00:00 Pengantar 20 Proyek Sumber Terbuka Teratas yang Sedang Tren di GitHub Baru-baru Ini 00:39 Gambaran Umum Mega Tren Utama yang Dieksplorasi Melalui 20 Proyek 01:23 Teknologi Agen untuk Kolaborasi Kerja dengan Mempekerjakan AI sebagai Anggota Tim 04:10 Solusi Memori Jangka Panjang Mengatasi Keterbatasan Memori Jangka Pendek Agen 05:50 Agen Penggunaan Komputer yang Mengontrol Desktop dan Antarmuka Secara Langsung 07:34 Generasi Video AI dan Visualisasi Data Spasial dengan Memanfaatkan Sumber Daya PC Lokal 08:53 Inovasi dalam Alat Pengembangan Inti Melalui Optimasi dan Pengurangan Beban Lingkungan Pengembangan 11:13 Proses Verifikasi Output AI dan Peningkatan Keamanan untuk Informasi Pribadi dan Izin 12:43 Virtualisasi Sumber Terbuka dan Inovasi Kustom yang Berawal dari Kesenangan dan Rasa Ingin Tahu 14:01 Kedatangan Era Agen Otonom, Refleksi tentang Peran Manusia, dan Kesimpulan Proyek Debugger RAD: EpicGames/raddebugger OpenRig: mvschwarz/openrig Agent Reach: Panniantong/Agent-Reach God's Eye View: bilawalsidhu/gods-eye-view Hindsight: vectorize-io/hindsight knowledge-work-plugins: anthropics/knowledge-work-plugins cua: trycua/cua RemoveMacAI: omlahore/RemoveMacAI bloodborne_pcbloodborne_pc: deadinside28/bloodborne_pc FreeVideo: FlashML-org/FreeVideo leviathan: elstongun/leviathan Agent Memory Repo: AgentMemoryRepo/agent-memory livenerf: ninjahawk/livenerf mygo: egoist/mygo OpenDots: CopilotKit/OpenDots clearai-dsh: Clearailhc/clearaidsh omnirush-gui: omnirush-ai/omnirush-gui garis rambut: lucasmarkes/hairline bops: OrgoAI/bops gdp-ts: rauchg/gdp-ts