Point the bundled engine at OpenRouter.
One key, many models, and the only route here that lets you split Plan, Edit and Vision across different providers.
OpenRouter is not a CLI engine like the other four. It is a provider for Lucid Train's own bundled engine: you paste an OpenRouter key in the app, it is written to ~/.lucid-train/config.toml, and requests go straight from your machine to OpenRouter. That gives you one key across many model vendors and lets you assign different models to the Plan, Edit and Vision roles. The same provider mechanism accepts local Ollama and any other OpenAI-compatible endpoint.
A provider, not a harness
The distinction matters for what you get. Claude Code, Codex, Cursor CLI and OpenCode each bring their own agent loop and you are running theirs. OpenRouter brings only a model endpoint, so the loop is Lucid Train's: its planner, its tool gating, its approval tiers. If you want the harness described everywhere else on this site rather than someone else's, this is the route that gives it to you.
Model routing across roles
Because the engine is Lucid Train's, it can send different roles to different models: a fast, cheap model to draft the plan, a strong coder for the diff, a vision-capable model to read a diagram back. That is not available when the engine is a third-party CLI, because those CLIs decide their own model routing.
Where the key lives
In ~/.lucid-train/config.toml on your machine. Requests go from your machine to OpenRouter and never through any server Lucid Train operates, which is the same guarantee that applies to every provider option including local models.
The alternative that costs nothing
The same provider slot accepts a local Ollama endpoint, which needs no key, sends nothing off the machine and works with no network at all. OpenRouter is the answer when you want frontier models without a per-vendor account; Ollama is the answer when you want no account at all. Any other OpenAI-compatible endpoint, including a self-hosted one, works through the same setting.
Provider options for the bundled engine
| Option | Key needed | Runs offline |
|---|---|---|
| OpenRouter | Your OpenRouter key | No |
| Local Ollama | None | Yes |
| Any OpenAI-compatible endpoint | Whatever it requires | If self-hosted |
Questions
Related
- Run Lucid Train on a Claude or ChatGPT SubscriptionNeither route asks you for an API key. One of them is a supported integration and one of them is interop; the difference matters.
- Run Lucid Train on Claude CodeThe same binary you already run in a terminal, driven from a window that also has your diagrams, your dependency audit and your dashboards in it.
- Run Lucid Train on the Codex CLIJSON-RPC to a local app-server, and the only engine besides Claude Code that can also drive the design canvas.
- Run Lucid Train on the Cursor CLICursor's headless agent, with the diagram canvas and the approval gate wrapped around it.
Last updated