Configuration

Lucid Train works with zero configuration. Anything you set is stored locally in ~/.lucid-train/config.toml and survives reinstalls · keys are never sent anywhere except the model provider you choose.

API keys

Run /apikey with no argument and paste your key into the masked prompt · it saves to your config automatically. Or pass it once:

shell

/apikey sk-or-v1-...

You can also export it before launching:

shell

$ export OPENROUTER_API_KEY=sk-or-v1-...

$ lucid-train

Config file

~/.lucid-train/config.toml · edits merge, so hand-editing one key never clobbers the rest:

~/.lucid-train/config.toml

model = "kimi-k2.7-code"

provider = "openrouter"

approvals = "auto"

max_iterations = 150

update_check = true

# Connect an MCP server

[mcp_servers.github]

command = "npx"

args = ["-y", "@modelcontextprotocol/server-github"]

Providers

Ollama (local): /provider ollama · fully on-device, free.
OpenRouter: /provider openrouter · Kimi, GPT, Gemini, Grok, DeepSeek with your key.
Self-hosted: /provider https://host:8000/v1 · any OpenAI-compatible endpoint.

Project memory

When working in a repo, the agent maintains .lucid/knowledge.md (architecture notes, gotchas, working commands) and .lucid/CHANGELOG.md, and reads them back in future sessions. Commit them to share context with your team.

Environment variables

Variable	Purpose
OPENROUTER_API_KEY	Default OpenRouter key for cloud models.
LLM_API_KEY / LLM_BASE_URL / LLM_MODEL	Pin any OpenAI-compatible endpoint (self-hosted vLLM, llama.cpp, SGLang, TGI).
LLM_FALLBACK_MODELS	Comma-separated failover list if the primary model is unavailable.
LUCID_TRAIN_HOME	Config directory (default ~/.lucid-train).
LUCID_UPDATE_URL	Override the update-check source (defaults to the PyPI JSON API).
LUCID_NO_UPDATE_CHECK	Set to disable the startup update check entirely.