AI Team OS plugin
A plugin for Claude Code and Codex: a shared task wall, project memory, reports and a dashboard that shows agents' work across sessions
- 374 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- Writes its own hooks into Claude Code and Codex settings and runs a local API server that reads session records and transcripts
- Stores decisions, notes and session messages in a local database, where secrets and project code may end up
- The tool set includes destructive operations such as deleting a project or team and restarting the API; a read-only profile can limit them
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/ai-team-osDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Writes its own hooks into Claude Code and Codex settings and runs a local API server that reads session records and transcripts
- Stores decisions, notes and session messages in a local database, where secrets and project code may end up
- The tool set includes destructive operations such as deleting a project or team and restarting the API; a read-only profile can limit them
This is third-party code. Review the repository files before installing.
What it does
AI Team OS adds a shared layer to Claude Code and Codex that keeps tasks, notes, decisions, reports and messages across sessions. One session ends and the next one gets the same records. The agents themselves stay native to their host, while AI Team OS keeps a log of who is busy with what and where work can resume. The plugin connects as an ai-team-os MCP server and gives the agent tools for project work. With them it reads the task wall and notes, writes progress and decisions, and searches earlier decisions. For Claude Code, hooks on session start, context compaction and subagent launch inject a startup briefing and the layer of user preferences. Codex connects through a separate adapter with its own hooks and trust approval. Memory has two layers. The direction layer stores preferences, corrections and design intent with character quotas and auditable invalidation. The task-note ledger is searched by keywords, including Chinese text. Data lives locally in the ~/.aiteam directory, and a web dashboard shows leaders, team members, tool activity and project totals. The channel_send, channel_read and channel_wait tools carry messages between sessions.
Who it is for. For developers who run several parallel Claude Code or Codex sessions and want one picture of the work and decisions that survive between sessions.
Good fit when
- Several Claude Code sessions work on a project at once and you need to know who took which task and how it ended
- Claude Code and Codex should share one project with common memory, notes and reports
- Preferences like "leave migrations alone" and user corrections should be stored apart from the chat and injected at session start
- You need a dashboard showing current agent work, tool activity and project totals
- An audit needs a read-only profile: the AITEAM_READONLY variable strips every write tool
Not a fit when
- You work in one short session with a single agent: the shared layer and dashboard do not repay the setup
- You want a very small tool set: the default profile registers about thirty tools, and the full set is much larger
- You work on Windows: the Codex adapter is validated on macOS and needs POSIX, and it was not validated on Windows
- You only need to remember preferences: a separate lightweight agent memory is simpler for that
Example request
Resolve this project in AI Team OS, read its task wall and task notes, and record progress and decisions thereLimitations
- Requires Python 3.11 or newer, uv and Claude Code or Codex with MCP support. The React dashboard needs Node.js 20 or newer and is optional.
- Install into the system Python rather than a project virtual environment, otherwise it works only in that project.
- Pick one path, the plugin or a source install. Using both loads the MCP server twice in every session.
- After the plugin install, Claude Code configures dependencies on first launch, and the time depends on the machine.
- The Codex adapter is validated on macOS and needs POSIX. Codex hooks are approved separately in its interface.
- Hooks and templates with tool restrictions are Claude Code specific and are not applied by Codex. Some restrictions work only in Claude Code versions that support them.
How to disable. The Claude Code plugin can be disabled with claude plugin disable or removed, but its global hooks stay in place, so look at the preview of leftovers first. The Codex adapter is removed with python3 scripts/codex_adapter.py uninstall --apply, and without the flag it only shows the plan. The database, session history and credentials are not deleted.
FAQ
How do I install AI Team OS in Claude Code?
Run pip install uv, then claude plugin marketplace add CronusL-1141/AI-company and claude plugin install ai-team-os. After a restart Claude Code configures dependencies on first launch. A source install exists as python3 install.py, but pick one of the two.
How do I connect Codex to AI Team OS?
Clone the repository, run python3 -m pip install -e ., then python3 scripts/codex_adapter.py install --api-url http://127.0.0.1:8000 and status. The adapter installs the MCP connection and hooks, makes backups and preserves unrelated settings. Hooks are approved in the Codex interface.
Where does AI Team OS keep its data?
The database, API port file, logs and install records live in ~/.aiteam, and AITEAM_DATA_DIR changes the path. Older versions kept data in ~/.claude/data/ai-team-os. The aiteam paths command shows which directories the current shell sees.
How do I reduce the number of tools the server loads?
The AITEAM_TOOLSETS variable selects module groups: the default profile covers tasks, teams, memory and reports, and default,ecosystem adds research tools. The AITEAM_READONLY variable keeps read tools only, which suits audits.
How is AI Team OS different from ordinary agent memory?
Besides decision memory it keeps a task wall, messages between sessions and a dashboard of leader and member activity. The host runs the agents, and AI Team OS replaces neither the scheduler, the permissions nor the agent lifecycle.
Discussion
No questions or reviews yet.
Related
PluginsAI Builder Club Skills
AI Builder Club — Skills
A Claude Code plugin marketplace to make a repo agent-ready and run compounding agent loops
A local dashboard for running many agents across machines: memory, code graph and agent-to-agent messaging
An operating layer for Claude Code and Codex: one /do entry point, project memory across sessions, safety hooks, cost telemetry and parallel agents
A self-improvement plugin for Claude Code, Codex and OpenCode: turns corrections into rules the agent follows in later sessions
Multi-agent orchestration for Claude Code: teams, roles, autopilot and task routing across skills and subagents
A long-term memory plugin and server for Claude Code and Codex: decisions and handoffs move between sessions and agents
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| AI Team OS (this card) | no | 374 | |
| oh-my-claudecode | no | 40,625 | multi-agent orchestration inside Claude Code, while AI Team OS stresses shared memory and a work log across sessions and hosts |
| Citadel | no | 927 | an operating layer for Claude Code and Codex with one /do entry point and project memory |
| sudocode | no | 296 | a memory and coordination layer where specs and issues live in git |
README in short
The README opens with the idea of a shared layer for Claude Code and Codex, then lists capabilities: session coordination, memory, tool-set control, the dashboard and safety rules. It continues with comparisons to similar projects, the architecture, three install paths, a description of data directories and removal instructions.