CluxMate: a Python coding agent for any OpenAI-compatible model
CluxMate
A coding agent with a Python core and three front-ends: a script CLI, a terminal TUI and an Electron app, for OpenAI-style APIs
- 140 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- The agent runs shell commands and edits files, and the sandbox is off in `yolo` mode
- Repository configuration (hooks, MCP, LSP) can spawn subprocesses if the directory is trusted
Install
Manual install
pip install .From the cloned r1c7/CluxMate folder. Python 3.12+ is required. Then run cluxmate.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The agent runs shell commands and edits files, and the sandbox is off in `yolo` mode
- Repository configuration (hooks, MCP, LSP) can spawn subprocesses if the directory is trusted
This is third-party code. Review the repository files before installing.
What it does
CluxMate is a coding agent that reads a codebase, plans changes, edits files, runs commands and answers questions. A single Python core powers three front-ends: a headless CLI for scripts and CI (`cluxmate -p "..."`), an interactive Textual TUI (`cluxmate`) and an Electron desktop app that talks to the core over JSON-RPC. The agent works with any model over an OpenAI-compatible API: DeepSeek, Qwen, GLM, OpenAI, OpenRouter, Ollama or your own endpoint, set through `base_url`. Every session is stored as an event log, so any request to the model can be replayed verbatim. Tools fall into four risk levels (`safe`, `write`, `dangerous`, `critical`) and four permission modes: `plan`, `default`, `acceptEdits`, `yolo`. File writes are guarded by a WriteFence, shell commands run in an OS-level sandbox and network requests pass an SSRF guard. There are skills, project memory in `AGENTS.md`, MCP servers with OAuth, lifecycle hooks, subagents, LSP and git worktrees.
Who it is for. Developers who want a coding agent with model choice and strict isolation: DeepSeek, Qwen, GLM or their own model with no tie to one vendor.
Good fit when
- An agent for scripts and CI with a one-shot command-line request is needed
- You switch between DeepSeek, Qwen, GLM, OpenRouter and local Ollama without changing tools
- Every request to the model must be replayed verbatim from the event log
- Shell commands should run in a sandbox and secrets such as `.env` and `~/.ssh` should be hidden from the model
- A repository needs a separate git worktree on the agent's own branch
Not a fit when
- You need a frontier model with ready subscriptions: CluxMate works through an API key and `base_url`
- You want a ready binary without Python: installation needs Python 3.12 or newer
- You want yolo mode without consequences: it disarms the sandbox
- The project is early: version 0.1.0, and interfaces and config format may change
Example request
Run cluxmate -p with the request to explain the session log design in this repository and show which tools the agent used.Limitations
Python 3.12 or newer is required, and the desktop app also needs Node.js 18 and npm. The agent needs an API key and the `base_url` of an OpenAI-compatible model. Version 0.1.0 is early, so interfaces and the config format may change.
`yolo` mode disarms the sandbox, and the other modes ask before dangerous commands. Configuration inside the repository's `.cluxmate/` counts as part of the repository and is read only after you explicitly trust the directory. Secret read protection is switched on separately, and by default the model sees files that are not under a denylist.
How to disable. Remove the package with `pip uninstall cluxmate`. Settings live in `~/.cluxmate/` and project settings in the repository's `.cluxmate/`: delete them by hand. `cluxmate worktree remove` deletes worktrees.
FAQ
How do I install CluxMate?
Clone the repository and run `pip install .` in it: the `cluxmate` command appears on your `PATH`. For the desktop app, also enter `desktop` and run `npm install`.
Which models are supported?
Any with an OpenAI-compatible API: DeepSeek, Qwen, GLM, OpenAI, OpenRouter, Ollama or your own endpoint through `base_url`. You can configure several and switch on the fly.
How do permission modes differ?
`plan` only reads. `default` asks about every write and dangerous command, `acceptEdits` accepts edits without asking, and `yolo` removes prompts and disarms the sandbox.
What is project trust?
Configuration in the repository's `.cluxmate/` (hooks, `mcp.json`, skills) can run commands, so it is read only after your consent. The `cluxmate trust status`, `trust add` and `trust deny` commands manage the state.
How do I connect an MCP server with OAuth?
Put the address in `mcp.json` with an `oauth` block, or log in with `cluxmate mcp auth linear`. `cluxmate mcp status` shows the state and `cluxmate mcp logout` deletes stored credentials.
Discussion
No questions or reviews yet.
Related
CLI toolsA Python terminal coding agent in the manner of Claude Code: reads the repository, edits files, runs commands, with memory, sub-agents and MCP
A Rust coding agent with a native GUI, a terminal mode, an ACP mode for Zed and an MCP server
Kode CLI is a terminal coding agent that reads AGENTS.md and CLAUDE.md, connects MCP servers and works with many models.
A terminal coding agent that picks a model for each request on its own and shows spend in real time
A terminal coding agent in Go: GLM, DeepSeek and local Ollama models, 59 built-in tools and MCP
A terminal coding agent in a single Go binary for DeepSeek and other OpenAI-compatible models, with a code graph and local OCR
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| CluxMate: a Python coding agent for any OpenAI-compatible model (this card) | no | 140 | |
| Kode CLI | no | 5,237 | a coding agent from the catalogue, while CluxMate has an event log and a two-layer sandbox |
| Aider | no | 50,089 | a pair programmer in the terminal, without worktrees or an OS-level sandbox |
README in short
The README covers the architecture of one core and three front-ends, replayable sessions, a completion audit, a two-layer sandbox, an SSRF guard and egress control. Then come project trust, worktrees, subagents, LSP, skills, MCP with OAuth, hooks, installation, usage and configuration.