AImanager CLI
A skill and the ai-collab tool for Codex CLI, Claude Code and API models to work on a project, plus file reading in ChatGPT
- 212 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- Participant agents edit files and run commands in the specified folder with explicit permission, and a folder path does not replace container isolation
- The account component works with Codex, Claude Code and Antigravity sign-ins and tokens, and ChatGPT reads files on your machines through MCP, so access mistakes and published ports expose data
- The selected text goes to API providers and their data retention rules apply to it
Install
- Works in
Manual install
python3 scripts/install-ai-collab.pyRun from the root of the repository clone, then ai-collab init and ai-collab doctor. Linux only, Python 3.8 or newer. For Claude Code set your directory with --skill-dir.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Participant agents edit files and run commands in the specified folder with explicit permission, and a folder path does not replace container isolation
- The account component works with Codex, Claude Code and Antigravity sign-ins and tokens, and ChatGPT reads files on your machines through MCP, so access mistakes and published ports expose data
- The selected text goes to API providers and their data retention rules apply to it
This is third-party code. Review the repository files before installing.
What it does
AImanager has several parts. Two matter for agents: the shared skill `multi-model-collaboration` and the `ai-collab` tool. From the current agent, such as Codex or Claude Code, they call other CLIs and API models to compare approaches, make edits, cross-review changes and continue each participant's conversation. The coordinator supplies context and permissions and verifies the result. CLI participants work in the specified folder, while API models get only selected text. The tool installs from the repository root with `python3 scripts/install-ai-collab.py`, followed by `ai-collab init` and `ai-collab doctor`. Tasks start with `ai-collab run --model <alias> --workspace <folder> --prompt <text>`, results come through `ai-collab wait` and `ai-collab result`, and conversations continue through `ai-collab reply`. The model aliases in the examples, such as `opus5.5` and `gemini4`, are set up by you. Four editable recipes cover independent review, planning and execution, research discussion and writing. The second part, LabContext, lets ChatGPT read authorized folders on your computer or server through MCP: directory trees, code, DOCX, XLSX, PPTX, notebooks and text PDFs. The third part, a web interface and a Tauri desktop app, shows sign-ins for Codex, Claude Code and Antigravity, quotas, usage history and session recovery.
Who it is for. Researchers and developers who work with Codex CLI, Claude Code and API models and want them to discuss and cross-check each other's work.
Good fit when
- A failing test should be reproduced in Codex CLI first and its edge cases checked in Claude Code
- You want to show an experiment design or a paper section to a second model and get notes on controls and alternative explanations
- After edits you need an independent check by another participant who reruns the tests
- ChatGPT should answer about project files on a server and you do not want to upload them each time
- You need quotas and sessions of Codex and Claude Code in one window and copyable resume commands
Not a fit when
- You need the tool on Windows or macOS without Linux: the published ai-collab executor currently requires Linux and Python 3.8 or newer
- You need an always-on scheduler: the project deliberately works without one
- You need container-level isolation for the code that runs: a workspace path is not such a boundary
- You need Antigravity as a collaboration participant: it has no dedicated adapter yet and support covers account management only
Example request
Work with opus5.5 on this project: let each of you propose a solution independently, discuss the key differences and implement the agreed changesLimitations
- The `ai-collab` tool works on Linux and requires Python 3.8 or newer. The other workspace components are not tied to Linux.
- Editing and code execution happen only after explicit permission. Starting collaboration from the web goes through the optional MCP bridge, where execution is off by default.
- Detailed documentation is mostly in Chinese, while the README exists in English and Russian too.
- Automatic extraction and updating of recipes from work outcomes is listed as planned and not yet implemented.
How to disable. Delete the skill from the agent's skills directory (the Codex one by default, or the one set by the `--skill-dir` flag) and the `ai-collab` file from `~/.local/bin`. For the LabContext part stop the Router and Tunnel services and revoke the connection in ChatGPT settings. Diagnostics and recovery are available through `labcontext status`, `labcontext doctor` and `labcontext repair`.
FAQ
How do I install ai-collab for Codex or Claude Code?
From the repository root run `python3 scripts/install-ai-collab.py`, then `ai-collab init` and `ai-collab doctor`. Add `~/.local/bin` to `PATH`. By default the skill goes into the Codex directory; for another agent set `--skill-dir`. Start a new session after installing.
How do I start a task on another model?
`ai-collab run --model opus5.5 --workspace /path/to/project --prompt "..."` returns a job ID. Read the result through `ai-collab wait ID --timeout 30` and `ai-collab result ID`, and continue the conversation with `ai-collab reply ID --prompt "..."`.
What does ai-collab doctor check?
Only configuration and connectivity, that is, whether the environment is ready. It does not judge whether tasks are solved correctly.
Can ChatGPT read my files without uploading them?
Yes, through the LabContext component and MCP, but only in authorized folders, with limits on size, depth and response volume. Read access grants no arbitrary execution. Default deployments listen on loopback only.
What does the opus5.5 alias in the examples mean?
It is an example alias from a personal configuration. It guarantees no model availability. The real mapping between aliases and providers is set up by you.
Discussion
No questions or reviews yet.
Related
CLI toolsShared infrastructure for Claude Code, Codex, Antigravity CLI and OpenCode: a task lifecycle, skills and sandboxes.
AI Agents Orchestrator
AI-Agents-Orchestrator
The AI Collaboration plugin hands a task between Claude Code, Codex, Cursor, Gemini CLI and Copilot via a shared SQLite history.
A tmux-based TUI that runs Claude Code, Codex and other agents in one list: status, quick prompts, worktrees and diff review
A plugin for Claude Code and Codex: a shared task wall, project memory, reports and a dashboard that shows agents' work across sessions
Claude Code skills that delegate UI to Gemini via agy and code to Codex, with cross-review and parallel git worktrees.
PromptOps keeps Claude Code, Codex and Gemini CLI side by side by project and gives a lead agent MCP to launch helpers
README in short
The page shows three scenarios: ChatGPT answers from project files, several models change and review code together, and one window manages accounts and quotas. Then come sample requests, installation by asking an agent, tables of modes and statuses, security boundaries and documentation links.