Skip to content

afumu/openteam

Folders and files

NameName
Last commit message
Last commit date

Latest commit

ย 

History

216 Commits
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 

Repository files navigation

OpenTeam

๐ŸŒ Language: English | ็ฎ€ไฝ“ไธญๆ–‡

A local-first AI team workspace that turns your existing ChatGPT, Claude, Gemini, DeepSeek, and Grok web sessions into a multi-agent discussion room.

OpenTeam is a Manifest V3 Chrome extension. It does not require model API keys and does not spend extra OpenAI, Claude, Gemini, DeepSeek, or Grok API tokens. Instead, it reuses the AI accounts you already have open in your browser, sends tasks to those web sessions, and gathers replies from different people and models into one shared team chat.

Use it for learning, research, and personal non-commercial experiments where one answer is not enough: product reviews, technical design reviews, content planning, personal decisions, multi-model comparison, and multi-step work that benefits from several AI roles thinking together.

๐ŸŒฑ Background

OpenTeam is a sister project to OpenLink, an earlier browser extension I built around using the browser itself as the working surface. While building OpenLink, I already wanted a companion extension that could reuse signed-in browser sessions to talk with large language models and compare answers across models, but that idea stayed on the shelf while other work got busy.

The idea became concrete after I met Lu (YUANLU007), a news editor. In her writing workflow, she often needs to compare several AI systems, inspect how different models answer the same question, and use multiple perspectives for factual checks. That matched the original OpenLink idea closely, so OpenTeam became its sister project: a local browser workspace for sending one task to multiple AI web sessions and gathering their responses in one discussion room.

OpenTeam group chat preview

๐Ÿ’ฌ Community

Join the OpenTeam WeChat community by scanning the QR code below. If the QR code has expired, add WeChat ID afumudev and include OpenTeam in your request so I can invite you to the group.

OpenTeam WeChat community QR code

โœจ Highlights

  • ๐Ÿšซ 0 API token workflow: reuse AI website sessions instead of calling model APIs directly.
  • ๐Ÿงฉ Multi-model discussion: coordinate Gemini, ChatGPT, Claude, DeepSeek, Grok, and other supported web sessions in one chat.
  • ๐Ÿง‘โ€๐Ÿซ Built-in advisor library: start with 38 expert and thinking-style advisor templates, or create your own people.
  • ๐Ÿ“ฃ Mention-based routing: use @person to ask one member, or @everyone to dispatch the same task to the whole team.
  • ๐Ÿ”„ Independent and collaborative modes: compare isolated perspectives first, then let members reference, challenge, and build on each other.
  • ๐Ÿ’พ Local-first storage: chats, people, messages, notes, highlights, and settings are stored in browser storage.
  • ๐Ÿค– Agent control CLI: optional openteamcli support lets local agents create chats, add roles, post tasks, and wait for replies.

๐Ÿงญ How It Works

Each OpenTeam member is bound to an AI website session. When you send a message, OpenTeam builds a prompt from the chat mode, member persona, referenced messages, and shared context. The extension then delivers that prompt into the member's iframe-backed AI page and listens for the reply.

team.html
  -> background service worker
  -> AI site iframe
  -> content script
  -> AI webpage reply
  -> OpenTeam message stream

Supported site types:

Site Common fit
Gemini Long context, research, multimodal material
ChatGPT General execution, tool-shaped workflows, fast iteration
Claude Long documents, review, structured writing, careful reasoning
DeepSeek Code, reasoning, Chinese-language work, cost-sensitive tasks
Grok Current-events research, alternative phrasing, X/Grok web-session workflows

These are practical defaults, not rankings. Actual quality depends on model versions, account plans, task type, and input material.

โš ๏ธ Disclaimer

OpenTeam is an unofficial project for learning, research, and personal non-commercial use only. It is not affiliated with, endorsed by, or supported by OpenAI, Anthropic, Google, DeepSeek, xAI, or any supported AI website.

OpenTeam interacts with AI websites through user-authenticated browser sessions and DOM automation. Website changes, account rules, rate limits, anti-abuse systems, or terms of service may affect whether it works. You are responsible for reviewing and complying with the rules, policies, laws, and regulations that apply to your use.

Do not use OpenTeam for commercial products, hosted services, paid workflows, bulk automation, spam, scraping, bypassing access controls, bypassing paid access, or any activity that violates third-party terms or rights.

Use it at your own risk. The maintainers are not responsible for account restrictions, service interruptions, data loss, legal issues, or other consequences arising from your use. This notice is not legal advice.

๐Ÿš€ Install

Option 1: Build the Extension From Source

Most users only need the browser extension:

git clone https://github.com/afumu/openteam.git
cd openteam
npm install
npm run build

After the build finishes, open chrome://extensions/, turn on Developer mode, click Load unpacked, select the generated dist/ directory, and then click the OpenTeam extension icon to open the team workspace.

Option 2: CLI + Agent Skill

Use this path if you want Codex, Claude Code, or another local agent to control OpenTeam chats. Install the CLI from npm, then install the agent skill:

npm install -g @afumu/openteamcli
npx skills add afumu/openteam --skill openteam-control
openteamcli daemon start
openteamcli doctor

The skills installer will ask which agent, scope, and install method to use. Use the GitHub command above for the public repository; if you are developing from a local checkout or unzipped source package, install the skill from the repository root instead:

npx skills add . --skill openteam-control

After installation, open the OpenTeam extension page and enable local agent control in settings. See OpenTeam CLI for more CLI details.

๐Ÿ› ๏ธ Development From Source

npm install
npm run dev

Build the extension and load the generated dist/ directory from chrome://extensions/ with Developer mode enabled.

Useful checks:

npm run typecheck
npm test
npm run build
npm run verify

npm run verify runs type checking, unit tests, and a production build.

๐Ÿ” Permissions and Privacy

OpenTeam is local-first, but it needs browser-extension permissions that are worth reviewing before installation:

  • storage: save chats, people, notes, settings, and local state.
  • tabs: locate and communicate with extension and AI-site tabs/frames.
  • alarms: schedule runtime maintenance tasks.
  • declarativeNetRequest: adjust response headers so supported AI websites can be embedded in the extension iframe workspace.
  • clipboardRead / clipboardWrite: support copy and clipboard-based interactions.
  • Host permissions: allow OpenTeam to work with supported AI websites and embedded web sessions.

OpenTeam does not provide cloud sync. Your AI conversations are still processed by the AI websites you use, under their own account rules, usage limits, privacy policies, and terms of service.

๐Ÿšง Current Limits

  • OpenTeam is Chrome / Chromium-first.
  • AI-site adapters depend on page DOM structure, so website redesigns can break prompt sending or reply capture.
  • The iframe workspace requires response-header changes through declarativeNetRequest, which makes the permission surface heavier than a normal popup extension.
  • Built-in famous-person advisor templates are prompt templates inspired by public ideas. They are not real people and should not be presented as real participation.
  • Medical, legal, financial, and other high-stakes outputs still require human judgment and qualified professional advice.

๐Ÿ—‚๏ธ Repository Layout

public/                 Chrome extension manifest, team page, styles, DNR rules
src/background/         service worker, command handlers, runtime routing
src/content/            AI-site content scripts, adapters, reply observation
src/group/              group data model, store, roles, prompts, mention parsing
src/teamPage/           OpenTeam workspace UI
packages/openteamcli/   local CLI and daemon for agent control
docs/                   design documents and assets

๐Ÿค Contributing

Issues and pull requests are welcome. Good starting areas include:

  • AI-site adapter fixes when target websites change.
  • Permission and privacy hardening.
  • Test coverage for group routing, prompt construction, storage, and UI flows.
  • Documentation, examples, and onboarding improvements.
  • New advisor templates or orchestration patterns with clear boundaries.

Before opening a pull request, run:

npm run verify

๐Ÿ“š Documentation

๐Ÿ“œ License

OpenTeam is released under the PolyForm Noncommercial License 1.0.0, using the SPDX identifier PolyForm-Noncommercial-1.0.0.

You may use, study, modify, and redistribute the project for non-commercial purposes only. Commercial use, commercial redistribution, hosted commercial services, paid workflows, and productized commercial use are not permitted without separate written permission.

About

A Chrome extension for orchestrating local AI-agent team workflows across web AI services.

Resources

License

Contributing

Stars

Watchers

Forks

Packages

 
 
 

Contributors