CodeLeveler CLI
A local coding agent with a choice of provider: DeepSeek, Kimi, OpenAI or Anthropic. Terminal and browser interfaces.
- 162 ★ on the repository
- Last commit
- Apache-2.0
Broad access
Why this level
- The agent edits files and runs commands and tests in the repository.
- The provider key is stored in settings on your computer.
Install
- Works in
Manual install
brew install dengmengmian/tap/levelerHomebrew for macOS and Linux x86_64; a `curl | sh` script and Windows archives also exist.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The agent edits files and runs commands and tests in the repository.
- The provider key is stored in settings on your computer.
This is third-party code. Review the repository files before installing.
What it does
CodeLeveler takes a task, reads code, edits files and runs tests in your local repository. You can work in the `leveler` terminal UI, in a browser with `leveler web` or with one command `leveler run "find the failing tests and fix them"`. Changes are visible as it works and `leveler resume` continues a session. Providers: BigModel, DeepSeek, Moonshot and Kimi, OpenAI, Anthropic and other OpenAI-compatible ones. Sessions stay on your machine while model requests go to your provider. The parallel mode `leveler run --parallel 2` keeps the permission profile, network restriction and step ceiling and prints jsonl. On Linux restricted commands need `bubblewrap` and on Windows `leveler.exe` and `leveler-confine.exe` sit together.
Who it is for. Developers who want a local coding agent with a choice of provider, permission limits and change review.
Good fit when
- Failing tests must be found and fixed with one `leveler run` command.
- Two solution attempts run in parallel and results are compared.
- Work runs on DeepSeek or Kimi models.
- A session resumes the next day with `leveler resume`.
Not a fit when
- You need a plugin or skill for an agent you already use: CodeLeveler is itself a coding agent.
- The Git working tree is dirty: changes are hard to review.
- Code must not go to the chosen model provider.
Example request
Find the failing tests in the project and fix them, showing changes before the commit.Limitations
A provider key is needed and `leveler login` sets it up. Start with a clean Git working tree so changes are easy to review, while `leveler doctor` shows configuration problems. The parallel run rejects shipping flags (`--commit`, `--push`, `--pr`, `--branch`, `--pr-base`), so ship the result separately. The install is a `curl | sh` script, so read it before running. The `--parallel` mode creates worktrees for candidates and cleans them after a run, but if cleanup fails the paths stay in warnings. Windows builds ship as archives and updates are done by hand.
How to disable. Remove the `leveler` binary (for example `brew uninstall leveler`) and the sessions and settings directory and revoke the provider key with the provider.
FAQ
Which models are supported?
BigModel, DeepSeek, Moonshot and Kimi, OpenAI, Anthropic and other services with an OpenAI-compatible interface.
Where are sessions stored?
On your computer. Only model requests go to the configured provider.
How do I install it on Windows?
Download the archive from Releases, verify the SHA-256 checksum, extract it and add the directory to `PATH`. Keep `leveler.exe` and `leveler-confine.exe` together.
What is needed on Linux?
Install `bubblewrap`, for example `sudo apt install bubblewrap`, before restricted commands.
How do I run two attempts at once?
With `leveler run --parallel 2 "task"`. Shipping flags are not accepted in this mode.
Discussion
No questions or reviews yet.
Related
CLI toolsTerminal coding agent in Go for DeepSeek, Kimi, GLM and OpenAI, built around prefix caching and confirmed writes and commands
A terminal coding agent in a single Go binary for DeepSeek and other OpenAI-compatible models, with a code graph and local OCR
A Python terminal coding agent in the manner of Claude Code: reads the repository, edits files, runs commands, with memory, sub-agents and MCP
Lightweight Python terminal coding agent: ReAct loop, file-based context, guarded bash and skills, works with DeepSeek and OpenAI
A terminal coding agent tuned for the deepseek-v4 model: deep thinking, reasoning effort control, Agent Skills and MCP
A coding agent for the terminal, browser and CI across 38 model providers, with cross-session memory, MCP, skills and hooks.
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| CodeLeveler (this card) | no | 162 | |
| Qwen Code | no | 29,777 | a terminal agent from the Qwen team; CodeLeveler adds a browser UI and parallel attempts with a permission profile |
| Nanocoder | no | 2,567 | runs on local models through Ollama; CodeLeveler connects Chinese and cloud providers |
README in short
The README presents a local coding agent: script and Homebrew install, Windows and Linux notes, first run, a table of core commands and a description of a parallel mode that keeps the permission profile and limits. Apache 2.0 license with a Chinese translation.