agent-zero

想让 agent 自己动手，又怕它乱动电脑？ Agent Zero 给 agent 一台独立 Linux 电脑，终端、浏览器、桌面应用沙箱跑，随时能接管，也能连本地文件。 适合想放手让 agent 干活的人，敏感文件先别接，权限看清楚。 你会给 agent 配独立电脑吗？ #Agent #沙箱 @grok agent0ai/agent-zero

Agent Zero v2.15: customizable welcome screen, chat search, per-action tool permissions, and security fixes Hey everyone, version 2.15 of A0 just landed. It's a big release, spread across a lot of different areas, so here is a short tour of the new features and the changes. Welcome Screen, now fully customizable - Reorder blocks with drag handles (mouse or touch) or the keyboard, and the order sticks - Dismiss the sections you don't use, bring them back anytime with "Show all dismissed" - A persistent puzzle-piece button opens the Plugin Hub right from the Welcome Screen Three new plugins: Welcome Chats, Welcome Files, Welcome Notes You won't find these in the changelog because they are plugins, not core. Click the puzzle piece on the Welcome Screen to find them. - Welcome Chats: find conversations and search message text from the Welcome Screen - Welcome Files: browse folders, filter files, and open them in Agent Zero from a responsive panel - Welcome Notes: movable, resizable notes for the Welcome Screen and chats, with Agent Zero, Paper, and Notebook themes They make the Welcome Screen more useful, but they are also meant as inspiration: if you have ever wanted to build your own panel, these are a good place to start. Search - Sidebar chat and message search with highlighted excerpts and keyboard navigation - Standard, Compact, and No search modes Tools are now plugins - wait, notify_user, search_engine, a2a_chat, scheduler, behavior_adjustment, and the scheduler Tasks UI are now standalone plugins - Per-action tool permissions: profiles can allow or deny individual operations across all 12 bundled multi-action tools, for example allow file reads but block writes, or list scheduled tasks but block deletion. The Agent Editor shows them as grouped, expandable controls - The four memory tools are merged into one memory tool with an action argument Web UI - Faster streaming: visible updates arrive roughly twice as often - The file browser is now per chat, with its own folder and history, plus a setting to keep it shared - Switch agent profile while a chat is running - Process steps render as plain text, so dollar signs no longer turn into math - Copy full execution commands while chat previews stay compact Agents - Fresh chats are no longer named after the greeting - Subordinate agents get the same response example as the main agent, which fixes the extra misformatting some subordinates had - Less redundant same-profile delegation: specialists handle matching work themselves - Agent Editor list loads 35% faster, with clearer prompt editing Migration - Pi Agent import (JSONL session versions 1 to 3) through Migrate Agents - Claude chat imports keep custom titles and tool-result blocks CLI - F6 chat and message search with highlighted excerpts - Main and Utility context-window settings in /models - Custom composer choices for plugins that span Web UI and connector - Redesigned branding and splash screen, faster startup and large chats Security - Same-origin isolation for the browser DOM helper, so cross-origin windows can no longer read page content or trigger DOM actions through it - Telegram and WhatsApp settings changes now require explicit administrators, so accepted senders can no longer lift their own tool restrictions or toggle plugins If you use Telegram or WhatsApp in A0, update and refresh your bots auth. Breaking changes - Telegram configuration is now global only. Existing project or profile-scoped Telegram configs will pause the bot on startup until you review and remove them - input_remote now also requires remote terminal permission (code_execution_remote) How to update - CLI: open your terminal and type a0 update - Web UI: open Settings > Check for updates > Review update Full changelog: github.com/agent0ai/agent-zero/releases/tag/v2.15 New here? Install guide: agent-zero.ai If you build a Welcome Screen plugin of your own, it would be great if you want to share it here. All it takes is "create <description-of-your-plugin> using a0-create-plugin skill under a0/skills that does this and that".

Agent Zero Alternative: How to Run a Private Agent on Your Mac or PC in 2026 (No Docker) --- title: "Agent Zero Alternative: How to Run a Private Agent on Your Mac or PC in 2026 (No Docker)" published: true description: "Off Grid AI is a finished app that works in your real apps with local models and asks first. Agent Zero runs in Docker and needs a model provider." tags: ai, privacy, productivity, opensource canonical_url: getoffgridai.co cover_image: getoffgridai.co --- Agent Zero gives an AI its own computer. Its GitHub README says it plainly: "Give your agent a full Linux computer." The agent gets a Linux desktop in Docker, a browser, a terminal and plugins. You install Docker, create an instance and choose a model provider. That is a great playground for builders. Most people want help in the apps they already use. Off Grid AI is a finished app that works on your own screen with local models, already knows your work, and asks before it sends anything. Get Pro for $79.99 once · Download Off Grid AI <figure> <img src="getoffgridai.co" alt="Off Grid AI Actions: a calendar invite, a Linear issue and a Gmail message waiting for approval, each showing where it came from." loading="lazy" /> <figcaption>Pending approvals in OGAD: each action shows where it came from and waits for you.</figcaption> </figure> Off Grid AI and Agent Zero side by side | | Off Grid AI | Agent Zero | |---|---|---| | How you install it | Download the app | A0 Launcher plus Docker, a script, or docker run | | Where the agent works | The apps on your own screen | A Linux desktop inside a Docker container | | Models | Local models you download in the app | Provider you configure, such as Ollama, LM Studio, OpenAI or Anthropic | | Before it acts | Mail, issues and docs wait for your approval (Pro) | Its safety notes tell you to review sensitive actions | | Memory | Your screens, meetings, mail and docs, on your disk (Pro) | Agentic RAG memory per project | | Phone | iPhone and Android apps | Not listed | | License | AGPL-3.0 | MIT | | Price | Free app; Pro $79.99 once or $7.99 a month | Free; model fees are separate | Checked 30 October 2026 against the Agent Zero README, its site and its releases. "Not listed" means the feature is not on Agent Zero's own pages. What setup looks like The README recommends the A0 Launcher. You download it for macOS, Windows or Linux, on x86 or ARM64. It checks Docker, creates instances, manages ports and connects to local or remote installs. There are two other ways. A terminal installer (curl on macOS and Linux, PowerShell on Windows) creates Dockerized instances. If Docker is already installed, you can run the image yourself. Then you open the Web UI and configure your model provider. Agent Zero's site lists many providers. Ollama and LM Studio cover local models. OpenAI, Anthropic, Google Gemini, DeepSeek, Groq, Mistral and OpenRouter cover the cloud. In Off Grid AI Desktop (OGAD), there is no Docker and no provider form. You open Models and download a model. The app shows which models fit your computer first. What runs without asking Agent Zero is honest about its power. Its README has a short Safety Model. It tells you to: - Keep it inside Docker or another isolated environment. - Not mount your whole home folder unless you understand the risk. - Grant host access and remote code execution only on machines you trust. - Review actions that touch accounts, money, production systems or private data. - Keep backups. So the container is the main limit, and the review is your job. Agent Zero also has Time Travel snapshots, which the README calls a practical safety layer for the workspace. It also says they do not replace Git or backups. Off Grid AI puts the check in the app. With Pro, Actions drafts replies, issues and docs through Gmail, Linear, Jira and Notion. Each one waits for Approve, Edit or Reject. God follows rules you set, such as "never send anything to Acme without asking". Your real apps, not a sandbox desktop Agent Zero's agent works inside its own Linux desktop. To reach your real files, you connect the A0 CLI as a bridge to your host machine. OGAD's Computer Use works in the apps you already use, on Mac, Windows and Linux. You watch each run and can Pause, Stop or Take Over. Tasks keeps the steps and screenshots. Web Use can work in your normal signed-in browser through the Off Grid AI extension, now in early access. Read how to automate a Linux desktop task with local AI. It already knows your work Agent Zero keeps memory, files and instructions per project. You fill it. Off Grid AI Pro fills its memory from your day. Your screens, meetings, mail and documents become one local memory on your disk. Ask "what did Sam say about the rollout?" and you get the answer with the source. God, your chief of staff, uses it for a morning brief, meeting prep, routines and scheduled tasks. When Agent Zero is the better pick Be fair to Agent Zero. It is free and MIT licensed. Its prompts, tools and settings are open to read and edit. It has 100+ community plugins, MCP servers, subagents and live editing of documents and spreadsheets. It runs on anything with Docker, from a $6 VPS or a Raspberry Pi to a GPU server. It shipped v2.14 on 13 October 2026, per its releases. If you want an agent with a full Linux machine to build on, Agent Zero is a strong pick. If you want an assistant for your own mail, meetings and apps, on your own models, that asks first, use Off Grid AI. Switch in ten minutes 1. Download OGAD for Mac (Apple silicon), Windows or Linux. 2. In Models, download a text model that fits your computer. 3. Get Pro and enter your key in the app. 4. Connect Gmail, Linear, Jira or Notion in Integrations. 5. Ask for a draft and approve it when it looks right. Your MCP servers can come with you. See how to connect a local AI to your apps through MCP. FAQ Is Agent Zero free? Yes. Agent Zero is open source under the MIT license. Its site does not list a price for the framework. You pay for any hosted model provider you connect. Does Agent Zero need Docker? Yes. Its README says Agent Zero runs wherever Docker runs. The A0 Launcher checks Docker and creates instances for you, and the terminal installer sets up Dockerized instances. Can Agent Zero use local models? Yes. Its site lists Ollama and LM Studio next to cloud providers such as OpenAI, Anthropic and Google Gemini. You configure the provider in its Web UI. How is Off Grid AI different from Agent Zero? Off Grid AI is a finished app with no Docker and no provider setup. It works in the apps on your own screen, builds a local memory of your work with Pro, and its actions in Gmail, Linear, Jira and Notion wait for your approval. What does Off Grid AI Pro cost? On 30 October 2026, Pro costs $79.99 once for lifetime access, a limited-time price down from $119, or $7.99 a month. One key covers up to 5 devices. Download Off Grid AI and get an agent that works in your own apps.

🔥 Браузер на автопилоте, агент с доступом к ОС и ещё 1 - находки GitHub #shorts #github #opensource Больше полезных находок каждый день - в Telegram: t.me Подборка интересных находок с GitHub за 15.10.2026: outcrawlai/scout, agent0ai/agent-zero, opengeos/RustGIS. Сохраняй, чтобы не потерять 💾 outcrawlai/scout: outcrawlai/scout Scout - open-source браузерный агент на TypeScript: LLM управляет браузером через CDP и выполняет веб-задачи сам. Пригодится тем, кто автоматизирует работу с сайтами и делает ИИ-агентов, которым нужен доступ к вебу. agent0ai/agent-zero: agent0ai/agent-zero Agent Zero - фреймворк для автономных ИИ-агентов на Python с веб-интерфейсом: агент работает в Linux-окружении, сам выполняет код и команды в терминале и использует их как основной инструмент. Подойдёт тем, кто хочет собрать расширяемого помощника под свои задачи и посмотреть, как устроен агент с доступом к системе. opengeos/RustGIS: opengeos/RustGIS RustGIS - открытая геоинформационная система, целиком написанная на Rust, для работы с геопространственными данными и снимками дистанционного зондирования. Будет полезна тем, кому нужна быстрая бесплатная альтернатива классическим GIS-программам, и тем, кто следит за Rust-экосистемой. #shorts #dev #github #useful #opensource гитхаб, гитхаб подборка, лучшее на гитхаб, новое гитхаб, гитхаб бесплатно, гитхаб ии, гитхаб мсп, гитхаб репы, гитхаб репозитории, лучшие репозитории, рейтинг гитхаб, github ai, best github, new github

Agent Zero v2.14: faster startup and chat switching, new TTS model, cost and better subagent controls, plus a Launcher update Hey everyone, v2.14 just landed. This one is mostly about the Core: optimization and stability, with a few new things on top, plus a Launcher update. preview.redd.it Faster startup and chat switching - Boot to first token time is down 35%, and opening a chat is 4x faster than v2.13 - Embedding warm-up now overlaps chat restoration and startup services New: Paradee Heart local speech - A 9MB text-to-speech model, added alongside Kokoro - English-only for now, think of it as a starting option - Default for unconfigured settings, existing voice selections are preserved Cost controls - Configurable context breakdown and usage rows - Selectable price currencies with cached exchange rates Subagents and chats - Directly delegated agents show their working state and can be stopped independently; parent cancellation propagates correctly, and nested children appear in the sidebar - Fixed disappearing chat and task lists when Sidebar Folders is disabled - Manual chat names stay intact under always-on automatic naming, and rejected custom names no longer disable automatic naming - Safer model preset editing: current presets reload before opening, so stale drafts cannot delete presets created elsewhere, and late selector responses cannot overwrite another chat's selection Models and providers - Improved prompt caching: cacheable prefixes are preserved in long Responses conversations, with cache routing for Venice and the Agent Zero API and configurable hints for custom providers - Better model compatibility: reasoning-field fallback, Claude prompt adaptation across third-party providers, corrected repeat detection for native tool calls - Nebius, Venice, Z AI Coding plan, and OpenRouter embeddings now route through LiteLLM instead of custom endpoints Launcher update - Configure saved Remote Instances: edit URLs, credentials, and Host access while keeping appearance settings - Host-specific certificate trust for Remote Instances with self-signed or private-CA certificates, with hostname and validity checks retained - Versions stay browsable and saved Remote Instance credentials stay available without a local runtime - Better Windows Docker Desktop and independent WSL Engine discovery, runtime selection, and Ubuntu setup when no ordinary WSL distro exists, with visible WSL setup progress - Launcher settings preserved during Windows upgrades and reinstalls, including recovery from destructive legacy uninstallers and interrupted installations How to update - CLI: open your terminal and type a0 update - Web UI: open Settings > Check for updates > Review update - Launcher: close and reopen the app, then click the Update button Full changelog: github.com/agent0ai/agent-zero/releases/tag/v2.14 New here? Install guide: agent-zero.ai If you try Paradee Heart, tell us how a 9MB voice holds up for you, and which languages you would want next. The model still sounds a bit robotic sometimes, but it's impressive for being tiny.