Kolosal CLI
A developer console AI assistant based on Qwen Code: code analysis and edits, git automation, OpenAI-compatible APIs, Hugging Face and image input
- 464 ★ on the repository
- Last commit
- Apache-2.0
Broad access
Why this level
- The console agent edits project files, runs commands and works with git at your request
- OpenAI-compatible service keys and the Hugging Face token live in the environment or `.env`, and requests go to an external model
- Token spend is hard to predict in advance
Install
Manual install
npm install -g @kolosal-ai/kolosal-ai@latestNeeds Node.js 20 or newer. Check with `kolosal --version`. There is also a script, `curl -fsSL https://raw.githubusercontent.com/KolosalAI/kolosal-cli/main/install.sh | bash`.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The console agent edits project files, runs commands and works with git at your request
- OpenAI-compatible service keys and the Hugging Face token live in the environment or `.env`, and requests go to an external model
- Token spend is hard to predict in advance
This is third-party code. Review the repository files before installing.
What it does
Kolosal CLI is a developer console AI assistant built on the Qwen Code codebase. You run the `kolosal` command in a project folder and hold a dialogue: ask it to explain the codebase structure, refactor a function, write tests for a module or add error handling to all database operations. The assistant works on large codebases beyond the context window of ordinary models. The model connects in two ways. Through an OpenAI-compatible API you set the `OPENAI_API_KEY`, `OPENAI_BASE_URL` and `OPENAI_MODEL` variables, and through Hugging Face the `HF_TOKEN` and `HF_MODEL` variables. If the input contains an image, Kolosal CLI switches to a vision-capable model by itself, with the mode set by the `vlmSwitchMode` option. The per-session token limit goes in `.kolosal/settings.json`. Beyond code analysis the assistant automates routine work: it groups the week's commits by feature, writes a changelog, finds and removes `console.log` calls and checks code for SQL injection. The `/help`, `/clear`, `/compress` (compresses history to save tokens), `/stats` and `/exit` commands work inside a session. It installs with one install command, through npm or a ready package for macOS and Debian or Ubuntu; for Android there is a Termux script.
Who it is for. For developers who want a Qwen Code style console agent that works with any OpenAI-compatible API or Hugging Face models.
Good fit when
- You need a console assistant that connects to your own OpenAI-compatible server at the `OPENAI_BASE_URL` address
- You want to plug in a Hugging Face model through `HF_TOKEN` and `HF_MODEL`
- You must get up to speed on someone else's codebase fast: structure, dependencies, endpoints and authentication methods
- You need a changelog built from the last seven days of commits
- The request includes a screenshot and the assistant will switch to a vision-capable model on its own
Not a fit when
- You are happy with Claude Code, Codex or Gemini CLI and do not want to switch tools
- You need models running with no external keys: the assistant requires a key for an OpenAI-compatible service or a Hugging Face token
- You need installation through Homebrew: the README marks it as coming soon
- You need an actively updated project: the last commit in the repository dates from December 2025
Example request
Explain this codebase architecture and find all API endpoints and their authentication methodsLimitations
The author warns that Kolosal CLI may make several API calls per cycle, so token use can run high, and promises optimization later. You need a key for an OpenAI-compatible service or a Hugging Face token, and the OpenAI API is unavailable from Russia. Homebrew installation is marked coming soon in the README. The npm route needs Node.js 20 or newer, and building from source needs cloning and `npm run build`. The installer runs through `curl | bash`, with `sudo` on Linux. The last commit in the repository was made in December 2025. The project is based on Qwen Code, under the Apache 2.0 license with a NOTICE file.
How to disable. Remove the package with `npm uninstall -g @kolosal-ai/kolosal-ai`. For .pkg and .deb packages use the system package manager. Settings sit in `.kolosal/settings.json`, and the `.env` file with keys must be deleted by hand.
FAQ
How do I install Kolosal CLI?
On macOS and Linux run `curl -fsSL https://raw.githubusercontent.com/KolosalAI/kolosal-cli/main/install.sh | bash`, with `sudo` on Linux if needed. Or through npm: `npm install -g @kolosal-ai/kolosal-ai@latest` on Node.js 20 or newer. Verify the install with `kolosal --version`.
How do I connect a model?
For an OpenAI-compatible API set `OPENAI_API_KEY`, `OPENAI_BASE_URL` and `OPENAI_MODEL` in the environment or a `.env` file. For Hugging Face set `HF_TOKEN` and `HF_MODEL`.
How does image support work?
Kolosal CLI notices an image in the input and switches to a vision-capable model. The mode is set by `experimental.vlmSwitchMode` with the values once, session and persist, and `visionModelPreview` enables preview support.
How do I limit token use?
Set `sessionTokenLimit` in `.kolosal/settings.json`, for example 32000. The `/compress` command compresses conversation history, and `/stats` shows the current session's data.
Does Kolosal CLI work on a phone?
The repository has README-termux.md and the install-termux.sh script for Android. Install details and troubleshooting are in docs/INSTALLATION.md.
Discussion
No questions or reviews yet.
Related
CLI toolsA Go coding agent for open models: it turns a chat into tasks on separate branches, checks and merges the result and runs all projects in one window
Local coding agent with a terminal UI: it plans, drafts several candidates, tests them in a sandbox and repairs, so compact open models can handle code
An open-source coding agent for VS Code, JetBrains and the terminal with hundreds of models to choose from
OpenAI's coding agent that runs locally in your terminal, with IDE and desktop versions
Orchestrator plugin for OpenCode and Codex CLI: multiple agents, hooks, built-in MCPs, ultrawork mode and hash-anchored edits
A teaching terminal coding agent in Python, built as a "nanoGPT for agents": read it, take it apart and fork it
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Repository stars | Main difference |
|---|---|---|---|
| Kolosal CLI (this entry) | no | 464 | |
| Qwen Code | no | 28,798 | the original Qwen Code project that Kolosal CLI is built on, with its own sign-in and settings |
| Gemini CLI | no | 108,615 | Google's console agent for Gemini models; Kolosal CLI targets OpenAI-compatible APIs and Hugging Face |
README in short
The README presents Kolosal CLI as an AI workflow tool for developers built on Qwen Code models, listing features and install routes: a quick script, npm, manual package download and a source build. It then covers launching, two authentication routes, setting the token limit and vision mode, and sample requests for code analysis, development and automation. It ends with session commands, keyboard shortcuts and links for contributors.