Lucid Train

Lucid Train is a systems-design-first coding harness that turns a repository into an architecture diagram and gives that diagram to a coding agent as a specification. It suits developers who want architecture to shape implementation and who are comfortable choosing local or hosted model providers. The free CLI is the practical starting point, while the paid desktop tier is aimed at users who need a native interface and broader design tooling.

Overview

Lucid Train connects repository mapping with code changes instead of treating diagrams as static documentation. Its agent emits a semantic graph, then the ELK layout engine calculates positions and orthogonal edge routes. That approach is useful when the structure of an existing codebase should inform planned work, particularly for larger systems where a visual map clarifies dependencies.

Proposed edits are shown as diffs, and commands are assigned risk tiers; catastrophic commands are refused. These controls make agent activity easier to review and reduce the chance of an obviously destructive action, but developers still need to assess the diff and approve appropriate changes.

Key features

• Architecture-to-agent workflow: A generated diagram can be handed to a coding agent as a specification, keeping system design and implementation in the same workflow. Mermaid support and diagram import also help when the starting point is an existing diagram rather than a freshly scanned repository.
• Portable diagram outputs: Diagrams are Excalidraw scenes that export to SVG and PNG, or can become Markdown specifications. SVG and PNG suit visual sharing; Markdown provides a text-based handoff that can sit alongside project documentation.
• Model flexibility: Local Ollama models, OpenRouter, other OpenAI-compatible endpoints and hosted cloud models are supported. This gives privacy-conscious users a local route and lets others keep an existing provider relationship.
• Connector ecosystem: The Connectors tab includes GitHub, Vercel, Railway and Grafana, and can install arbitrary stdio MCP servers. The important boundary is that MCP connections are stdio-only: remote HTTP and SSE servers cannot connect directly.
• Local handling and credentials: Repository files are processed locally, while prompts and code go to the selected provider—or nowhere for local models. The maker says it collects no telemetry, analytics, prompts, code or usage data from local or bring-your-own-key sessions. API keys remain in the Lucid Train configuration directory or OS keychain rather than being sent to the maker.

Pricing

Plan	Price	What you get
CLI	Free forever	Full agent, local models or your own API keys, open source, and no account requirement
Lucid Train	$2.50 per month, billed $30 a year	Everything in CLI plus a native app, Design, Diagram, Audit and Dashboards; 7-day trial with no card up front

The CLI is the strong value choice for developers who mainly want repository-aware agent work and are happy with a command-line workflow. It gives up the native desktop experience and the additional Design, Diagram, Audit and Dashboards tools. The paid plan fits users who want those tools in one desktop application; its annual billing is inexpensive, but the seven-day trial is brief for evaluating a broader workflow. A desktop license covers up to three machines, and desktop sign-in is required for licensing.

Platforms

Lucid Train runs on Linux, macOS and Windows. The CLI is open source and needs no account, while the desktop application uses sign-in for licensing. That combination suits mixed-device developers, although anyone needing a web-only workflow or remote HTTP/SSE MCP connectivity should look elsewhere.

Who it's for

This is a good fit for developers and architects who want a repository map to guide an agent, especially teams balancing local models, bring-your-own API keys and hosted services. It is less suitable as a general collaborative whiteboard, a purely hosted diagram service, or a setup centered on remote MCP servers that expose HTTP or SSE endpoints.

Pros and cons

Pros

• Connects semantic repository diagrams directly to agent specifications, making architecture useful during implementation.
• Diff-based edits and risk tiers improve reviewability and refuse catastrophic commands.
• Supports local and multiple remote model routes, with local processing and locally stored credentials.
• Exports to Excalidraw, SVG, PNG and Markdown, covering visual and text workflows.
• Free CLI has the full agent and no-account access for experimentation.

Cons

• MCP is limited to stdio, excluding direct HTTP and SSE connections.
• The desktop license is capped at three machines and requires sign-in.
• The paid plan is billed annually at $30, so the low monthly equivalent is not a month-to-month term.
• The product is specialized around coding-agent and repository architecture work rather than broad diagram collaboration.

Alternatives

For a broader canvas with web, desktop, API and self-hosted options, consider Creately; its free plan includes unlimited canvases and collaborators but limits each canvas to 45 items. Nimbalyst is the better choice when MIT-licensed desktop and iOS apps and no feature limits on its Individual plan matter more than an architecture-to-agent workflow.

Choose OpenCharts if a web or macOS project workspace with 500K free AI credits, exports and Drive storage is the priority. InfraSketch suits users seeking a free infrastructure-diagram product across Linux, macOS, web and Windows. ArchitectureDiagram.ai is a simpler web/API option, though its free plan allows two diagrams per month and watermarked exports.

ClearWork is for paid, web-based workspace management with read-only users, rather than a free coding harness. Archway is worth choosing for hosted AI diagram generation with seven diagram types and a free monthly allowance. Capci fits teams that want API and MCP access alongside a free lifetime generation allowance.

For category browsing, see AI Diagram Generators and AI Wireframe Generators.

Verdict

Choose Lucid Train if you want repository architecture to become an actionable specification for a coding agent, with model-provider flexibility and a genuinely useful free CLI. Look elsewhere if you need direct HTTP or SSE MCP servers, a browser-first collaboration suite, or desktop licensing beyond three machines.