OpenSeek CLI
A MoonBit coding agent for DeepSeek, Kimi and GLM over an OpenAI-compatible API that runs without cloning the repository.
- 107 ★ on the repository
- Last commit
- Apache-2.0
Broad access
Why this level
- The agent runs scripts and processes and edits files in the working directory.
- Requests and code go to the model provider using the user's key.
Install
Manual install
moonx moonbitlang/openseek --helpNo install: `moonx` downloads and caches a prebuilt wasm binary. The MoonBit toolchain is required.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The agent runs scripts and processes and edits files in the working directory.
- Requests and code go to the model provider using the user's key.
This is third-party code. Review the repository files before installing.
What it does
OpenSeek is built from pure data modules, HTTP transport, agent orchestration and a CLI. The `run`, `serve`, `mcp` and `sessions` subcommands run one task, start a host server, validate MCP configuration and list sessions. The agent uses eleven local tools: `mbtx` for scripts and process launching without a shell, `edit`, `multi_edit`, `write`, `remove`, `plan`, `goal`, `finish` and others. The model is selected by key: `DEEPSEEK`, `KIMI` or `GLM`. Claude- and Cursor-style `mcp.json` files load through `--mcp-config`, and skills are read from `.openseek/skills` and `~/.openseek/skills`. Running through `moonx moonbitlang/openseek` fetches a prebuilt wasm binary from mooncakes.io.
Who it is for. MoonBit developers and anyone who wants a terminal agent on the Chinese DeepSeek, Kimi or GLM models instead of Anthropic or OpenAI.
Good fit when
- An agent should run on a project with one command, no install or clone.
- A task should run on DeepSeek, Kimi or GLM using their keys.
- A ready `mcp.json` from Claude or Cursor should plug in unchanged.
- An agent session should be watched live in the browser with `--inspect`.
Not a fit when
- You need an interactive interface: the terminal UI is a separate `openseek_tui` binary.
- OpenAI or Anthropic models are required.
- MCP resources and prompts are needed: the agent uses tools only.
Example request
Run openseek run in the project directory with the Kimi model and ask for a four-sentence description of the repository structure.Limitations
The author calls the project a small foundation rather than a finished product. A key from one of the providers (`DEEPSEEK`, `KIMI` or `GLM`) is required and requests go to their servers. The agent runs scripts and edits files in the working directory, so use it on a version-controlled project. MCP resources and prompts are not consumed, only tools. Building from source needs the MoonBit toolchain and `just`. The project is young, so the command interface may change.
How to disable. Remove the `moonx` cache and the `.openseek` directories in projects and the home folder. Unset provider keys from the environment.
FAQ
How do I choose a model?
Set `DEEPSEEK`, `KIMI` or `GLM`. For Kimi and GLM pass `--model`; the default is `deepseek-flash`.
Do I have to install anything?
Not to run it. `moonx moonbitlang/openseek` downloads and caches a prebuilt wasm binary.
How do I connect MCP?
Pass `--mcp-config` with a Claude- or Cursor-style file. Several files merge in order.
How can I watch a session?
Add `--inspect`: the agent starts a viewer and prints a link. It cannot be combined with `--no-session`.
Is there an interactive mode?
It lives in the separate `openseek_tui` module. The root package is for headless runs.
Discussion
No questions or reviews yet.
Related
CLI toolsAn MCP server that runs DeepSeek or your own local model as a sub-agent in Claude Code and Codex with its own tool loop.
A 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
A terminal coding agent built on Codex, tuned for low-cost open models like Kimi K3, DeepSeek and Qwen
Kimi Code is Moonshot AI's terminal coding agent: it reads the project, edits files and runs commands. Installs with one script, no Node.js needed.
Self-hosted AI agent in a single Go binary: terminal, web UI, chat apps and desktop all run on any compatible model
mi is agentic coding in 30 lines: a loop, two tools and a model; works with any OpenAI-compatible API and runs in Docker.
README in short
Running without a checkout, the monorepo layout, a package table, CLI commands, MCP, skills and verified documentation.