LUCID-TRAIN(1)General Commands Manual
Documentation
Install the CLI or the desktop app, sign in, point it at your models, and go. The same agent core drives both.
Quickstart
Install the CLI, launch it in any repo, then sign in to unlock the cloud models:
$curl -fsSL https://lucidtrain.com/install.sh | sh$lucid-train$lucid-train login
Prefer a window? Download the macOS desktop app — Code, Diagram, Audit, Dashboards and Connectors in one place. Credentials live in the OS keychain.
Full install channels — pip, pipx, uv, Homebrew, winget, cargo — are on the installation page.
Modes
Three modes, switched with a slash command at any point in a session:
| /agent | Full tools. Edits files and runs commands. The default. |
| /plan | Propose-only. A reviewed plan; nothing executes until you approve. |
| /research | web_search and read_url only — cited answers, no shell. |
The rest of the palette is in the command reference.
Your models, your machine
| Local · free | Run fully offline with Ollama — gpt-oss, qwen3-coder, deepseek-r1. No key, no telemetry. |
| OpenRouter · BYO key | 15+ models — GPT, Gemini, Grok, DeepSeek V4, Qwen, GLM, Kimi, MiniMax. |
| Cloud · no keys | Sign in with Google for hosted DeepSeek V4 Pro, GLM 5.2 and Kimi K2.7 Code. See plans. |
| Self-hosted | Point it at any OpenAI-compatible endpoint: vLLM, llama.cpp, SGLang, TGI. |
Sizing guidance, the full catalogue and failover are on the models page; keys and the config file are under configuration.
Next
- Installation — every channel, by OS.
- Configuration — keys,
config.toml, project memory. - Commands — the full slash-command reference.
- Grafana debugging — go from a firing alert to a merged fix.