lucidtrain.com

mx

aspmx.l.google.com

alt1.aspmx.l.google.com

alt2.aspmx.l.google.com

alt3.aspmx.l.google.com

alt4.aspmx.l.google.com

ns

ns28.domaincontrol.com

ns27.domaincontrol.com

caa

txt

v=spf1 include:dc-aa8e722993._spfm.lucidtrain.com ~all

google-site-verification=IHx7f20z_5r8wpx_WDgw0OioZ_Y-TNURauVh5MoGykk

dmarc

v=DMARC1; p=quarantine; adkim=r; aspf=r; rua=mailto:dmarc_rua@onsecureserver.net;

User-Agent: *
Allow: /
Disallow: /dl/

User-Agent: GPTBot
Allow: /
Disallow: /dl/

User-Agent: ChatGPT-User
Allow: /
Disallow: /dl/

User-Agent: OAI-SearchBot
Allow: /
Disallow: /dl/

User-Agent: ClaudeBot
Allow: /
Disallow: /dl/

User-Agent: Claude-User
Allow: /
Disallow: /dl/

User-Agent: Claude-SearchBot
Allow: /
Disallow: /dl/

User-Agent: anthropic-ai
Allow: /
Disallow: /dl/

User-Agent: PerplexityBot
Allow: /
Disallow: /dl/

User-Agent: Perplexity-User
Allow: /
Disallow: /dl/

User-Agent: Google-Extended
Allow: /
Disallow: /dl/

User-Agent: Applebot-Extended
Allow: /
Disallow: /dl/

User-Agent: DuckAssistBot
Allow: /
Disallow: /dl/

User-Agent: cohere-ai
Allow: /
Disallow: /dl/

User-Agent: Meta-ExternalAgent
Allow: /
Disallow: /dl/

Host: https://www.lucidtrain.com
Sitemap: https://www.lucidtrain.com/sitemap.xml

# Lucid Train

> Lucid Train reads a codebase and draws its architecture, then hands that
> diagram to a coding agent as a specification and returns reviewable diffs. It runs as a
> native desktop app (macOS, Windows) and as a single 6 MB Rust binary with a terminal UI.
> 
> The distinguishing facts, since they are what the comparisons usually get wrong:
> 
> - Diagrams are not placed by the language model. The agent emits a semantic graph of
>   nodes, edges and tiers with no coordinates; the ELK layout engine computes positions
>   and orthogonal edge routing. Overlapping boxes are therefore structurally impossible
>   rather than merely unlikely.
> - It is local-first. It runs fully offline against local Ollama models, with no API key
>   and no telemetry, or against any OpenAI-compatible endpoint using your own key.
>   Prompts go straight to the provider you choose and never through Lucid Train.
> - It is a desktop application, not a web app. Source code does not leave the machine.
> - It can drive a coding CLI you already have installed, as your own login: Claude Code
>   (claude -p), Codex (codex app-server), Cursor CLI (cursor-agent acp) and OpenCode
>   (opencode serve) are spawned as local processes. There is no Gemini CLI support.
>   Design mode runs on Claude Code or Codex only.
> - The Audit tab runs installed scanners (osv-scanner, npm audit, cargo audit,
>   pip-audit, snyk, socket) over dependency manifests across nine ecosystems, and also
>   scans the repository's own MCP configuration for unpinned npx servers, remote or
>   plaintext endpoints and secrets in env. It is not a source-code scanner: no SAST,
>   no SBOM, no container scanning, and it opens no pull request.
> - The CLI is free. The desktop app is $3/month or $30/year, with a 7-day trial. Model
>   usage is billed by your provider, never by Lucid Train.
> 
> Lucid Train is an independent product with no affiliation to Lucid Software, Lucidchart
> or Lucidspark.
> 
> WHAT IT IS NOT. Stated here so a summary does not have to guess:
> 
> - Not a browser tool. There is no web app, no share link, no live co-editing and no
>   cloud sync. Diagrams are files; git is the sharing mechanism.
> - Not a Mermaid or PlantUML tool. The agent can write either into a file the way any
>   coding agent writes text, but nothing renders, validates or round-trips them.
> - Not an LLM tracing or evaluation platform. For that use Langfuse, LangSmith, Arize
>   Phoenix or Braintrust. Lucid Train queries Grafana, Prometheus and Loki, which is a
>   different and smaller thing.
> - Design mode outputs self-contained HTML pages, landing pages and web app UIs, with
>   all styling inline. It does not emit React or Vue components, editable vector files,
>   or native iOS, Android or desktop application screens. The device buttons are
>   preview widths (100%, 768px, 390px), so the output is responsive web, not native.
> - Not an editor. No inline completion, no debugger, no extension ecosystem.
> - MCP support is stdio only. Remote HTTP and SSE servers cannot be connected.
> - There is no first-party Jira or Confluence integration. You can add Atlassian's own
>   stdio MCP server yourself; native support is on the roadmap.

## Product and docs

- [Home](https://www.lucidtrain.com): What Lucid Train is, and the download.
- [Pricing](https://www.lucidtrain.com/pricing): CLI free, desktop app $3/month or $30/year.
- [Free](https://www.lucidtrain.com/free): The CLI is free and open source; the desktop app has a 7-day no-card trial.
- [Releases](https://www.lucidtrain.com/releases): Downloads for macOS and Windows with SHA-256 checksums.
- [Docs](https://www.lucidtrain.com/docs): Quickstart: install, sign in, point it at your models.
- [Installation](https://www.lucidtrain.com/docs/install): pip, pipx, uv, cargo, curl, Homebrew and winget.
- [Configuration](https://www.lucidtrain.com/docs/configuration): API keys, config file, providers, context windows.
- [Commands](https://www.lucidtrain.com/docs/commands): Full slash-command reference for the terminal UI.
- [Models](https://www.lucidtrain.com/docs/models): Local Ollama, OpenRouter, or any OpenAI-compatible endpoint.
- [Grafana debugging](https://www.lucidtrain.com/docs/grafana): Query Loki and Prometheus, then drive a fix in the repo.
- [Engines](https://www.lucidtrain.com/docs/engines): Drive the Claude Code, Codex, Cursor CLI or OpenCode binary you already have.
- [Design mode](https://www.lucidtrain.com/docs/design): Generate landing pages and web app UIs as self-contained HTML.
- [Audit](https://www.lucidtrain.com/docs/audit): Scan dependency manifests and the MCP config your agent launches servers from.
- [MCP](https://www.lucidtrain.com/docs/mcp): Connect stdio servers, or run Lucid Train as one.
- [Troubleshooting](https://www.lucidtrain.com/docs/troubleshooting): The failures that look like bugs and are not.
- [Terms of Service](https://www.lucidtrain.com/terms): Terms of Service.
- [Privacy Policy](https://www.lucidtrain.com/privacy): Private by default.

## Templates

Each template is a real diagram produced by the same layout engine the app runs, not a picture of one.

- [System Design Templates](https://www.lucidtrain.com/templates)
- [URL Shortener System Design](https://www.lucidtrain.com/templates/url-shortener): The canonical read-heavy system: roughly a hundred reads for every write, and a redirect that has to be fast enough that nobody notices it happened.
- [Chat Application System Design](https://www.lucidtrain.com/templates/chat-application): Long-lived connections change everything: the hard part is not storing messages, it is knowing which of your servers is holding the socket you need to write to.
- [Payment System Design](https://www.lucidtrain.com/templates/payment-system): The only system on this list where being approximately right is indistinguishable from being wrong, and where the provider, not you, holds the truth.
- [RAG Pipeline Architecture](https://www.lucidtrain.com/templates/rag-pipeline): Almost every RAG system that disappoints is failing at retrieval, not generation, and the architecture is what decides whether you can tell.
- [E-commerce Platform Architecture](https://www.lucidtrain.com/templates/e-commerce-platform): Four services with four completely different consistency requirements, which is the whole reason this is not one application.
- [Notification Service Design](https://www.lucidtrain.com/templates/notification-service): The architecture is straightforward. The reason this needs to be one service is that the rules about when not to send are the hard part, and they have to live somewhere every sender passes through.
- [Distributed Rate Limiter Design](https://www.lucidtrain.com/templates/rate-limiter): Every design here is a trade between how accurate the limit is and how much latency you are willing to add to every single request to achieve it.
- [Observability Stack Architecture](https://www.lucidtrain.com/templates/observability-stack): The architecture is mostly about cost control. Collecting everything is technically easy and financially ruinous, so the interesting decisions are all about what to throw away.
- [Authentication Service Design](https://www.lucidtrain.com/templates/authentication-service): Stateless tokens make verification free and revocation hard, and that single tradeoff explains most of the components in this diagram.
- [Event-Driven Microservices Design](https://www.lucidtrain.com/templates/event-driven-microservices): Distributed transactions do not exist here, so every consistency guarantee you want has to be rebuilt out of events, retries and compensation.
- [LLM Inference Service Design](https://www.lucidtrain.com/templates/llm-inference-service): GPUs are the budget, so almost every decision here is about keeping them busy without letting the queue destroy tail latency.
- [CI/CD Pipeline Architecture](https://www.lucidtrain.com/templates/ci-cd-pipeline): Two things decide whether people trust a pipeline: how fast it is, and whether a red build actually means something is broken.
- [Ride Sharing System Design](https://www.lucidtrain.com/templates/ride-sharing-app): Every driver streams their position every few seconds whether or not anyone is riding, which makes this one of the few consumer systems where writes dominate reads.
- [Video Streaming System Design](https://www.lucidtrain.com/templates/video-streaming-service): Around ninety-nine percent of the bytes never touch your servers, which means most of this design is about preparing files so the CDN can do the work.
- [Social Media Feed System Design](https://www.lucidtrain.com/templates/social-media-feed): Fan-out on write is the right answer until someone with ten million followers posts, and the entire design exists to handle that exception.
- [Web Crawler System Design](https://www.lucidtrain.com/templates/web-crawler): The engineering problem is not fetching pages quickly. It is fetching them politely, at scale, without visiting the same content a thousand times.
- [Search Autocomplete System Design](https://www.lucidtrain.com/templates/search-autocomplete): This is a latency problem wearing a search problem's clothes: past about a hundred milliseconds users have already typed the next character and your response is worthless.
- [File Storage Service Design](https://www.lucidtrain.com/templates/file-storage-service): The design goal is to keep file bytes away from your servers entirely, so that what you run is a metadata service that happens to be about files.
- [Ticket Booking System Design](https://www.lucidtrain.com/templates/ticket-booking-system): A named seat can be sold exactly once, and ten thousand people want it at the same instant. Everything here follows from that.
- [Food Delivery System Design](https://www.lucidtrain.com/templates/food-delivery-app): Three sides to coordinate and one deadline that nobody controls: the kitchen. Dispatch quality matters more here than any amount of throughput.
- [Multiplayer Game Backend Design](https://www.lucidtrain.com/templates/multiplayer-game-backend): Latency is the product. Every architectural decision here is subordinate to keeping the round trip between a player and the simulation short.
- [Multi-Tenant SaaS Architecture](https://www.lucidtrain.com/templates/saas-multi-tenant-app): One database serving every customer, and exactly one bug standing between that and showing one customer another customer's data.
- [Data Warehouse ETL Pipeline Design](https://www.lucidtrain.com/templates/data-warehouse-etl-pipeline): Land the raw data first and transform it later, because the transformation you want in six months is not the one you would write today.
- [IoT Telemetry Platform Design](https://www.lucidtrain.com/templates/iot-telemetry-platform): Assume every device is offline, on a bad network, running firmware from two years ago, and cannot be recalled. The architecture follows from that.
- [API Gateway System Design](https://www.lucidtrain.com/templates/api-gateway): One entry point, so authentication, rate limiting and observability are implemented once instead of in every service.
- [Feature Flag Service Design](https://www.lucidtrain.com/templates/feature-flag-service): Evaluation has to be local and instant, because a flag check sits in the hot path of code that would otherwise not make a network call at all.
- [Webhook Delivery System Design](https://www.lucidtrain.com/templates/webhook-delivery-system): You are making requests to servers you do not control, which are frequently slow, sometimes wrong, and occasionally gone.
- [Audit Logging System Design](https://www.lucidtrain.com/templates/audit-logging-system): An audit log that can be edited is not an audit log, which makes this one of the few systems where the absence of features is the design.
- [Vector Search Service Design](https://www.lucidtrain.com/templates/vector-search-service): The failure mode here is not an error. It is recall dropping slowly as the index grows, with nothing in your metrics saying so.
- [AI Agent Orchestration Design](https://www.lucidtrain.com/templates/ai-agent-orchestration): The orchestrator owns the loop, and almost every hard problem here is about knowing when to stop.

## System design

The diagram can cause the code, not just describe it.

- [AI System Design](https://www.lucidtrain.com/system-design)
- [What Is a Systems-Design-First Coding Agent?](https://www.lucidtrain.com/system-design/systems-design-first-coding-agent): The architecture is the input, not the write-up.
- [AI System Design Tools, Compared](https://www.lucidtrain.com/system-design/ai-system-design-tools): Three different products share one search term.
- [From System Design to Code](https://www.lucidtrain.com/system-design/system-design-to-code): What has to be true for a drawing to become a build.
- [Architecture-Aware Coding Agents](https://www.lucidtrain.com/system-design/architecture-aware-coding-agent): Understanding a system is not the same as reading a file.
- [AI Software Architecture Tools](https://www.lucidtrain.com/system-design/ai-software-architecture-tool): Three jobs, three very different reliabilities.
- [A Coding Agent That Draws Diagrams](https://www.lucidtrain.com/system-design/coding-agent-with-diagrams): One model of the codebase, used twice.

## Harness engineering

Agent equals model plus harness. The harness is the half you can actually change.

- [Harness Engineering](https://www.lucidtrain.com/harness)
- [What Is a Coding Harness?](https://www.lucidtrain.com/harness/what-is-a-coding-harness): Everything in the agent that is not the model. The part you can actually change.
- [Harness Engineering vs Context Engineering](https://www.lucidtrain.com/harness/harness-engineering-vs-context-engineering): One is a component of the other, and confusing them is why teams optimise the wrong thing.
- [Coding Agent vs Harness](https://www.lucidtrain.com/harness/coding-agent-vs-harness): Agent equals model plus harness. The harness is the half you control.
- [Is Claude Code a Harness?](https://www.lucidtrain.com/harness/is-claude-code-a-harness): Both, and the confusion is in the question rather than the answer.
- [Is GitHub Copilot a Harness?](https://www.lucidtrain.com/harness/is-github-copilot-a-harness): Two products under one name, and only one of them is a harness.
- [Choosing a Coding Harness](https://www.lucidtrain.com/harness/best-coding-harness): The honest answer is that it depends on constraints, and here are the constraints that actually decide it.
- [Coding Harness for Local LLMs](https://www.lucidtrain.com/harness/coding-harness-for-local-llm): Three requirements, and most harnesses fail at least one of them.
- [Harness Design Patterns](https://www.lucidtrain.com/harness/harness-design-patterns): Three labs, three vocabularies, one shared concept between them. Here is the mapping.
- [Context Management in a Coding Harness](https://www.lucidtrain.com/harness/harness-context-management): Every harness works on a small task. Context management is what separates them at the limit.
- [Verification in a Coding Harness](https://www.lucidtrain.com/harness/verification-and-acceptance): The agent cannot mark its own homework, and there is now published evidence rather than just a suspicion.

## Engines

The binary is already on your machine and already signed in. Nothing is proxied and no credential passes through us.

- [Coding Engines Lucid Train Can Drive](https://www.lucidtrain.com/engines)
- [Run Lucid Train on Claude Code](https://www.lucidtrain.com/engines/claude-code): The 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 CLI](https://www.lucidtrain.com/engines/codex-cli): JSON-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 CLI](https://www.lucidtrain.com/engines/cursor-cli): Cursor's headless agent, with the diagram canvas and the approval gate wrapped around it.
- [Run Lucid Train on OpenCode](https://www.lucidtrain.com/engines/opencode): The open-source option, and the only engine that talks HTTP rather than stdio.
- [Run Lucid Train on OpenRouter](https://www.lucidtrain.com/engines/openrouter): One key, many models, and the only route here that lets you split Plan, Edit and Vision across different providers.
- [Run Lucid Train on a Claude or ChatGPT Subscription](https://www.lucidtrain.com/engines/run-on-your-subscription): Neither route asks you for an API key. One of them is a supported integration and one of them is interop; the difference matters.

## Audit

Your packages are audited by everyone. The programs your agent launches are audited by almost no one.

- [Dependency and Agent Supply Chain Audit](https://www.lucidtrain.com/audit)
- [MCP Server Security Scanning](https://www.lucidtrain.com/audit/mcp-server-security): An MCP server is a program with your agent's permissions. Most people install one without reading what it launches.
- [Dependency Vulnerability Audit](https://www.lucidtrain.com/audit/dependency-audit): Real scanners, run locally, with an agent that can then do the upgrades and show you the diff.
- [Offline Dependency Audit](https://www.lucidtrain.com/audit/offline-security-audit): For the repositories that are the reason your team cannot use a hosted scanner.
- [Security Audit Inside a Coding Agent](https://www.lucidtrain.com/audit/audit-in-a-coding-agent): Scanning where the code is, rather than reporting to somewhere the code is not.
- [Software Supply Chain Audit](https://www.lucidtrain.com/audit/supply-chain-audit): The packages you import are audited by everyone. The programs your agent launches are audited by almost no one.

## SRE

Nothing is exported to us. It reads the Grafana you already run, with your credential, on your machine.

- [Grafana and Observability in a Coding Agent](https://www.lucidtrain.com/sre)
- [Connect Grafana to a Coding Agent](https://www.lucidtrain.com/sre/connect-grafana): The tools are Grafana's own MCP server, not a reimplementation of the Grafana API.
- [Chat With Your Grafana Dashboards](https://www.lucidtrain.com/sre/chat-with-dashboards): Every answer names the query it ran, so you can check it rather than trust it.
- [From a Production Symptom to a Code Change](https://www.lucidtrain.com/sre/production-symptom-to-code): Not an automated root cause engine. One window that holds both halves of the investigation.

## Design

Running on the Claude Code or Codex CLI you already have, with your repository open beside it.

- [AI Design Mode for Web Pages](https://www.lucidtrain.com/design)
- [AI Landing Page Design](https://www.lucidtrain.com/design/ai-landing-page-design): A real page you can open and click, not a picture of one.
- [Wireframe to High Fidelity](https://www.lucidtrain.com/design/wireframe-to-high-fidelity): A finished-looking mockup gets reviewed as a design. A rough one gets reviewed as a structure.
- [Click to Edit an AI Design](https://www.lucidtrain.com/design/click-to-edit-design): The difference between iterating and rerolling.
- [Design With Your Claude Code or Codex Login](https://www.lucidtrain.com/design/design-with-claude-code): No second subscription for the design tool, because it is running the one you already pay for.
- [Design to Code](https://www.lucidtrain.com/design/design-to-code): Not generated components you then have to rewrite. A reference, and an agent that knows your codebase.
- [Responsive Preview at Mobile Width](https://www.lucidtrain.com/design/responsive-web-design-preview): Stated plainly because the buttons imply more than they do: these are viewport widths, not platforms.

## Glossary

What each term means, where it came from, and where it stops being useful.

- [Software Architecture Glossary](https://www.lucidtrain.co