Quickstart
Lucid Train is a single-binary coding agent for your terminal. It runs frontier open-source models locally via Ollama, or any model through OpenRouter. This guide gets you from zero to your first task in a minute.
1. Install
The recommended install is the pip wheel · it ships the compiled binary, no Python or Rust toolchain needed:
shell
$ pip install lucid-train
Prefer something else? See all install methods · pipx, uv, cargo, curl | sh, Homebrew, and winget.
2. Launch
shell
$ lucid-train
On first run you'll see a setup screen that detects your hardware. You have two paths · and you need no API key for the local one:
- Local & free: run
/download qwen3:4b(Lucid Train installs Ollama for you if it's missing), then/model qwen3:4b. - Cloud: run
/apikeyand paste your OpenRouter key (stored locally), then/model kimi-k2.7-codeor any model.
3. Give it a task
Just describe what you want. The agent reads the repo, plans, edits files (shown as diffs), runs commands, and verifies its work:
shell
› fix the failing auth test and run the suite
Headless / CI
Run a single task non-interactively and exit with a status code:
shell
$ lucid-train exec "add a healthcheck endpoint" -m gpt-oss:20b
Modes
/agent· full tools; edits files and runs commands (default)./plan· proposes a reviewed plan; nothing executes until you approve./research· web search + read only; cited answers, no shell.
Next steps
- Installation · every method, per-OS.
- Configuration · keys, config file, env vars.
- Commands · the full slash-command reference.
- Models · local, OpenRouter, and self-hosted.