Coding agent

An AI coding agent that runs on your machine.

Same agent core in a native desktop app and a 6 MB Rust binary, running whichever model you choose, including one on your own laptop.

Download for macOS

v0.1.33 · .dmg · Apple Silicon & Intel

Signed & notarized by Apple · opens without a Gatekeeper warning

sha256 698955a0187bc039f4c74f5d05a9f10fbb27376a45788a0a241d1326b73873c7

Download for Windows instead

$curl -fsSL https://lucidtrain.com/install.sh | sh

Design first, then implement

The distinguishing feature is not the coding loop, which most agents do competently now. It is that a turn can start from a diagram. Draw or generate the architecture, and one action converts the canvas into a markdown specification listing every component, relationship and note, then starts a coding turn with it. The agent implements against the drawing rather than against a paragraph describing the drawing.

Any model, including local

Any OpenAI-compatible endpoint works: local Ollama models entirely offline, OpenRouter for the open catalogue, or your own key with any provider. Different roles can use different models, so a fast one drafts the plan while a stronger one writes the diff and a vision model reads your diagrams. Prompts go straight to whichever provider you picked and never through us.

Propose, then apply

The default is propose-only: you see the plan and the diffs and approve them before anything is written. When you trust a task, the same agent flips to applying automatically. An agent that rewrites a repository without showing you first is a tool you use once.

More than a coding tab

The desktop app carries the coding agent, the diagram canvas, a dependency audit, a Grafana and Prometheus surface and connectors, all driven by the same agent core, with a command palette and multiple project workspaces running their own sidecars concurrently.

How it compares in this category

	Lucid Train	Typical AI coding tool
Model choice	Any OpenAI-compatible endpoint	Vendor's models
Fully offline	Yes, with local models	No
Code leaves your machine	No, with a local model	Yes
Visual architecture surface	Yes	No
Diagram as specification	Yes	No
Interface	Desktop app and terminal	Editor extension or terminal
Price	CLI free; desktop $3/month	Subscription, usually higher

The CLI is free

shell

$ curl -fsSL https://lucidtrain.com/install.sh | sh

FAQ

Questions

Which models does it support?

Any OpenAI-compatible endpoint: local Ollama, OpenRouter with the full open catalogue, or hosted models with your own key. Context windows are tracked and conversations compact before overflow.

Can it run fully offline?

Yes, with local Ollama models. No API key, no telemetry, nothing leaves the machine, and credentials live in the OS keychain.

How is this different from Cursor or Claude Code?

The visual design surface and model freedom. Neither of those has a diagram canvas that can act as the specification for a coding turn, and neither runs against arbitrary local models.

What does it cost?

The CLI is free and open source. The desktop app is $3/month or $30/year with a seven-day trial. Model usage is billed by your provider, or is free if you run locally.

Local LLM Coding Agent
For when the constraint is not cost but that the code cannot leave the building.
Diagram-Driven Development
The diagram stops being documentation produced after the fact and becomes the instruction the code is written from.
Terminal Coding Agent
One binary, no runtime, no browser. Free and open source.
AI Assistant for Observability
Worth being precise about what this is: it drives the observability tools you already run. It is not an LLM tracing platform, and if that is what you searched for, the tools below are the right ones.