chesto

Command-line AI agent powered by Kimi K3 — the strongest open-weight model. Chat in your terminal; it runs real system tasks (shell commands, file read/write) on your machine.

Like Claude Code runs on Anthropic's best model and Codex runs on OpenAI's best, chesto runs on the best open model.

Single file. Zero dependencies. Node ≥ 18.

Install

git clone https://github.com/jinyuanwong/chesto-cli.git
cd chesto-cli
ln -sf "$(pwd)/chesto.mjs" /opt/homebrew/bin/chesto   # or any dir in your PATH

Use

export CHESTO_API_KEY=sk-...   # get one at https://platform.moonshot.ai

chesto                      # interactive chat
chesto -p "do something"    # one-shot task
chesto --yolo               # skip per-command confirmation

Test the agent loop for $0 (no API key)

A local mock Kimi API server lets you verify the full loop — streaming, tool calls, real command execution — without spending anything:

node mock-server.mjs &
CHESTO_BASE_URL=http://localhost:11435/v1 CHESTO_API_KEY=mock chesto

Config

| env | default | notes | |---|---|---| | CHESTO_API_KEY | — | Moonshot API key | | CHESTO_BASE_URL | https://api.moonshot.ai/v1 | point to mock / self-hosted / any OpenAI-compatible backend | | CHESTO_MODEL | kimi-k3 | use kimi-k2.5 for ~5× cheaper development |

Implementation notes

• kimi-k3 requires reasoning_content to be preserved across tool calls — handled.
• kimi-k3 does not accept sampling params (temperature etc.) — omitted.
• Tool outputs are truncated at 50k chars to protect the context window.
• Commands ask for confirmation before running; --yolo disables that.

Roadmap

• Human-in-the-loop tools: connect the Chesto task marketplace (public discovery: agent-card.json · skill.md · llms.txt) so the agent can call real human actions, not just APIs.
• npm package (npm install -g chesto).

中文

命令行 AI Agent，底层是最强开源模型 Kimi K3。终端输入 chesto 对话，可直接执行本机系统任务。

chesto                      # 交互对话
chesto -p "帮我..."         # 一次性任务
chesto --yolo               # 不逐条确认命令

零成本本地测试（不花一分钱验证 agent 循环）：

node mock-server.mjs &
CHESTO_BASE_URL=http://localhost:11435/v1 CHESTO_API_KEY=mock chesto

省钱开发：CHESTO_MODEL=kimi-k2.5（约 1/5 价格）。

License

MIT