OpenAI Dots 发布一周，GitHub 冒出多个开源复刻

morph @morpphhhaw原推发布于 2026.10.07

在 X 上打开原推

OPENAI SHIPPED DOTS LAST WEEK. GITHUB ALREADY HAS TWO OPEN-SOURCE VERSIONS - AND EIGHT REPOS THAT REBUILD THE REST OF THE STACK Not ten chatbot wrappers Persistent computers, browser control, long-term memory, inbox workers, scheduled jobs, app connections and approval gates The pieces required to build a Dot outside ChatGPT are already public 1. OpenDots - probably the closest public reconstruction of the full Dots product Persistent agent computers, Spaces, browser control, calls, Slack, background work and approval cards. ↳ http://github.com/CopilotKit/OpenDots 2. Open Dot - a local Mac version of the same idea Each Dot gets a browser, workspace and memory. It supports schedules, event triggers, voice calls, encrypted credentials and human approval before risky actions. ↳ http://github.com/composio-community/open-dot 3. Meepo - a local Dot written in Rust Runs as a daemon and connects models to email, calendar, browser, GitHub, Slack, Discord and iMessage through more than 75 tools. ↳ http://github.com/leancoderkavy/meepo 4. Assistant - an autonomous worker that lives inside its own email inbox Queued jobs, recurring reminders, persistent memory, deep research, calendar access and a shared document workspace. ↳ http://github.com/rb81/assistant 5. Chatty - a browser-based agent for running small-business operations Reads email, manages a GTD task system, joins calls and wakes itself up on a schedule to check calendars, tasks and incoming work. ↳ http://github.com/WWilson1017/chatty 6. Bantz - a local-first desktop operator for Linux Email, calendar, browser automation and more than 70 tools, with a confirmation firewall around destructive actions. ↳ http://github.com/miclaldogan/bantz 7. OpenAI Agents SDK - the official runtime underneath a custom Dot Sessions, tools, handoffs, guardrails, tracing and durable human approval flows without inventing your own agent loop. ↳ http://github.com/openai/openai-agents-python 8. Browser Use - the computer-control layer Gives an agent a programmable browser so it can research sites, operate web apps and work through logged-in sessions. ↳ http://github.com/browser-use/browser-use 9. Mem0 - the memory layer Stores user preferences, decisions and useful context so every background run does not begin from zero. ↳ http://github.com/mem0ai/mem0 10. Composio - the connection layer More than 1,000 toolkits with authentication, tool search, context management and sandboxed execution for agent workflows. ↳ http://github.com/ComposioHQ/composio The useful split: > want a complete Dot → OpenDots or Open Dot > want it completely local → Meepo, Assistant or Bantz > want a small-business operator → Chatty > want to build your own → Agents SDK + Browser Use + Mem0 + Composio The interesting part of Dots is not the avatar It is the combination of persistent state, a computer that stays available, tools connected to real work and a hard boundary around actions that still require you