Skip to content
SPECIFICTION
Get listed

Lucid Train

lucidtrain.com

From $2.50/mo Free plan WindowsmacOSLinux
Visit lucidtrain.com
Spec score 8.2 #15 of 34 in AI Wireframe Generators

What it is

Lucid Train is a desktop coding harness that draws architecture diagrams from a codebase and passes the diagram to a coding agent as a specification. The agent emits a semantic graph, and the ELK layout engine positions diagram elements and routes edges. Diagrams are Excalidraw scenes that can export to SVG and PNG or become Markdown specifications; Mermaid support and diagram import are also listed. Proposed edits appear as diffs, and commands use risk tiers, with catastrophic commands refused. Model options include local Ollama, OpenRouter, other OpenAI-compatible endpoints, and hosted cloud models. Repository files are processed locally; prompts and code go to the selected provider, or nowhere when using 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 are stored locally or in the OS keychain. Connectors include GitHub, Vercel, Railway, and Grafana, plus arbitrary stdio MCP servers; remote HTTP and SSE servers cannot connect directly. The free CLI is open source. The native app costs $2.50 per month, billed $30 a year, and includes a 7-day free trial.

Spec sheet

Product
Lucid Train draws architecture diagrams from a codebase and hands the diagram to a coding agent as a specification.Read 5 Oct 2026 · lucidtrain.com
Intended users
The product is described as a systems-design-first coding harness for people who want to map a repository and use that diagram to guide a coding agent.Read 5 Oct 2026 · lucidtrain.com
Integrations
The Connectors tab lists GitHub, Vercel, Railway and Grafana and can install arbitrary stdio MCP servers.Read 5 Oct 2026 · lucidtrain.com
Support
The maker lists [email protected] for questions about its terms and privacy policy.Read 5 Oct 2026 · lucidtrain.com
Agent workflow
Edits are presented as diffs, and commands use risk tiers with catastrophic commands refused.Read 5 Oct 2026 · lucidtrain.com
Architecture diagrams
The agent emits a semantic graph and the ELK layout engine computes diagram positions and orthogonal edge routing.Read 5 Oct 2026 · lucidtrain.com
Credentials
API keys are stored locally in the Lucid Train configuration directory or OS keychain and are not sent to the maker.Read 5 Oct 2026 · lucidtrain.com
Data handling
The software processes repository files locally, and prompts and code go directly to the chosen model provider or nowhere when local models are used.Read 5 Oct 2026 · lucidtrain.com
Desktop license limit
A desktop license covers up to three machines, and desktop sign-in is required for licensing.Read 5 Oct 2026 · lucidtrain.com
Diagram exports
Diagrams are Excalidraw scenes that export to SVG and PNG and can also be turned into a Markdown specification.Read 5 Oct 2026 · lucidtrain.com
MCP limitation
MCP connections use stdio only; remote HTTP and SSE servers cannot be connected directly.Read 5 Oct 2026 · lucidtrain.com
Model support
Lucid Train supports local Ollama models, OpenRouter and other OpenAI-compatible endpoints, plus hosted cloud models.Read 5 Oct 2026 · lucidtrain.com
Telemetry
The maker says it does not collect telemetry, analytics, prompts, code or usage data from local or bring-your-own-key sessions.Read 5 Oct 2026 · lucidtrain.com

Plans and prices

CLI Free Free · forever Full agent · local models or your own API keys · open source · no account needed Read 5 Oct 2026 · lucidtrain.com
Lucid Train $2.50 a month billed $30 a year Everything in the CLI in a native app · Design, Diagram, Audit and Dashboards · local models or your own API keys · 7-day free trial · no card up front Read 5 Oct 2026 · lucidtrain.com

Every Lucid Train plan and price →

Strong points and limits

Strong points

  • Turns repository architecture into agent specifications
  • Exports diagrams as SVG, PNG, or Markdown
  • Supports local Ollama and hosted model providers
  • API keys remain local or in the OS keychain

Limits

  • MCP connections support stdio only
  • Desktop license covers up to three machines
  • Native app requires desktop sign-in for licensing

Who it suits

Lucid Train suits systems-design-first developers who want to map a repository and use the diagram to guide a coding agent. It may also suit users who prefer local models or their own API keys.

Lucid Train in depth

Read the full write-up

Lucid Train turns a codebase into an architecture diagram that can guide a coding agent. It is best suited to developers who want systems design to shape repository work and are comfortable choosing a model or running one locally. The free CLI covers the core agent workflow; the paid desktop app adds a broader set of design and diagram tools.

Overview

Lucid Train connects repository mapping to implementation rather than treating architecture diagrams as an end in themselves. Its agent emits a semantic graph, and the ELK layout engine calculates positions and routes orthogonal edges. That makes the product a focused fit for planning changes against a codebase, not a substitute for a general-purpose visual workspace.

Edits are presented as diffs, and commands are assigned risk tiers, with catastrophic commands refused. This makes agent proposals easier to review and adds a useful safeguard, but the developer still needs to judge the changes before accepting them.

Key features

Diagrams as coding specifications: A generated architecture diagram can be handed to a coding agent as a specification. This keeps system structure in the implementation workflow rather than leaving it as disconnected documentation.

Flexible diagram formats: Diagrams are Excalidraw scenes that export to SVG or PNG and can also become Markdown specifications. Mermaid support and diagram import make the workflow useful beyond newly generated repository maps.

Model choice: Lucid Train supports local Ollama models, OpenRouter, other OpenAI-compatible endpoints and hosted cloud models. Local processing can keep prompts and code away from external providers; using a hosted provider means those prompts and code go to the provider selected by the user.

Connectors and MCP: The Connectors tab includes GitHub, Vercel, Railway and Grafana, and users can install arbitrary stdio MCP servers. The stdio-only connection model rules out directly connecting remote HTTP and SSE servers, a meaningful constraint for teams whose integrations depend on those transports.

Pricing

The free CLI costs nothing, forever. It includes the full agent, local models or the user’s own API keys, is open source and requires no account. This is the strongest fit for developers who want the core workflow without paying for a native desktop app.

Lucid Train costs $2.50 a month, billed $30 a year. It includes everything in the CLI in a native app, plus Design, Diagram, Audit and Dashboards. A 7-day free trial is available with no card up front. The paid plan makes sense for users who need those desktop tools; the free CLI gives up that native-app feature set.

A desktop license covers up to three machines and requires desktop sign-in. API keys are stored locally in the Lucid Train configuration directory or OS keychain and are not sent to the maker. The maker says it collects no telemetry, analytics, prompts, code or usage data from local or bring-your-own-key sessions.

Platforms

Lucid Train runs on Linux, macOS and Windows. Its desktop license limit of three machines may matter to developers who work across more devices.

Who it's for

Choose Lucid Train if you want to map a repository and use that architecture to direct coding-agent work, especially if you value model choice, local processing, reviewable diffs and risk-tiered commands. It is less suitable if your workflow depends on remote HTTP or SSE MCP servers, or if you want a diagramming tool without a coding-agent focus.

Pros and cons

Pro: Architecture maps can become coding-agent specifications, tying repository structure to implementation decisions.

Pro: The free, open-source CLI includes the full agent and works with local models or personal API keys without an account.

Pro: Diff-based edits and risk tiers make proposed changes easier to inspect, with catastrophic commands refused.

Con: MCP support is limited to stdio, excluding direct connections to remote HTTP and SSE servers.

Con: The desktop license requires sign-in and covers at most three machines.

Alternatives

For a broader diagramming canvas, consider Creately: its free plan includes unlimited canvases and collaborators, with 45 items per canvas and one fold. For web-based collaboration centered on read-only access, ClearWork includes up to 10 read-only users on every core plan, while its Solo plan costs $200 a month.

Capci is a paid alternative with a free plan of 50 lifetime generations shared across posts, diagrams and animations, plus one brand. If you want a diagram-focused web workflow, Mermaid Chart has a free Basic plan with three diagrams, limited AI and one team and project; its Plus plan is $12 a month billed annually.

For open-source desktop and iOS apps, Nimbalyst offers MIT-licensed apps and source code on GitHub. OpenCharts may suit users looking for a freemium tool with 500K AI credits on its free plan. For a browser extension and web-based diagram workflow, Splotch offers a free plan with one project, 10 diagrams and two-way sync.

Archflow is another freemium, web-based option, with a free plan that includes one project and three diagrams per project.

Browse AI Diagram Generators or AI Wireframe Generators for more tools in adjacent categories.

Verdict

Lucid Train is a strong fit for developers who want repository architecture to guide coding-agent work: it combines a specification-oriented diagram workflow with model flexibility and a free full-agent CLI. Look elsewhere if remote HTTP or SSE MCP connectivity is essential, or if you need a general-purpose diagramming workspace rather than a coding harness.

Verdict

Lucid Train connects repository diagrams with a coding-agent workflow, with local processing and several model options. Its stdio-only MCP support and desktop licensing limit are worth checking before choosing it.

Where Specifiction places it

Alternatives to Lucid TrainAll alternatives →