Lucid Train is a local-first coding harness that generates architecture diagrams from repositories and turns them into specifications for an AI coding agent. It is available as a desktop app or a free, open-source Rust CLI.
Lucid Train is a local-first developer tool that generates architecture diagrams from source code and uses those diagrams as specifications for AI-assisted coding. It can survey a repository, produce a semantic graph of components and relationships, and hand the resulting architecture to its coding agent.
The product is offered as a desktop application and as a small Rust CLI. The desktop app combines coding-agent sessions with Diagram, Audit, Dashboards, and Connectors surfaces. The CLI provides the same agent core in a terminal-oriented workflow.
The agent reads a repository on the local machine and emits a semantic graph of components, tiers, and relationships. ELK computes the positions and orthogonal edge routing instead of relying on language-model-generated coordinates.
Generated diagrams are Excalidraw scenes that can be exported to SVG or PNG and edited in Excalidraw clients. A one-click action turns the canvas into a Markdown specification covering components, relationships, and narrator notes.
Run models locally through Ollama, bring an API key for an OpenAI-compatible endpoint or OpenRouter, or use the hosted model options described in the documentation. The site says prompts are sent to the provider selected by the user.
The CLI and agent support full tool use, propose-only planning, and cited research mode. Plan mode allows a reviewed plan before execution, while risk-tiered commands and reviewable diffs provide checkpoints for edits.
The desktop app can hold several repositories, with each project maintaining its own agent sidecar and history. It also includes a command palette, editor launching for VS Code, Cursor, or Zed, voice input, and image attachments for vision models.
The desktop app includes dependency checks using osv-scanner plus npm, cargo, and pip auditing, MCP configuration auditing, Grafana access for dashboards and telemetry, and support for stdio MCP servers and SKILL.md packs.
A new engineer can start from a diagram generated from the existing codebase instead of reconstructing the system solely from scattered files and documentation.
Teams can inspect the architecture of older systems or code produced across multiple AI coding sessions, then create an editable diagram and Markdown specification from the current repository.
Reviewers can base discussions on a diagram derived from the implementation. The generated specification can provide structured material for a design document or review workflow.
Teams can use repository-derived diagrams to examine the current system and make intermediate migration states explicit before asking the coding agent to implement changes.
Organizations that cannot send source code to a browser can run the workflow locally with Ollama. The site describes local execution as offline, with no API key required for that setup.
The agent surveys a repository and emits a semantic graph of components, tiers, and relationships. The ELK layout engine then computes positions and orthogonal edge routing, so the language model does not decide the box coordinates.
The site states that repository code is read on the user’s machine and that fully offline operation is available with local Ollama models. For external providers, prompts go to the provider configured by the user; credentials are stored in the operating system keychain.
Diagrams are Excalidraw scenes that can be exported to SVG and PNG and remain editable in Excalidraw clients. The desktop app can also turn a canvas into a Markdown specification listing components, relationships, and notes.
They use the same agent core. The CLI is a terminal-focused Rust binary with modes, plans, diffs, and token accounting, while the desktop app adds Code, Diagram, Audit, Dashboards, and Connectors in one window.
The CLI is free and open source with no account required. The desktop app has a seven-day free trial and is listed at $30 per year, or $2.50 per month when billed yearly. Model usage is separate: users can run local models or pay the provider associated with their own API key.
Traffic data is for reference only.
talkcody.com
TalkCody is a free, open-source AI coding agent for macOS, Windows, and Linux, with multiple model providers, local storage, and offline workflows.
opencode.ai
OpenCode is an open-source AI coding agent for terminal, IDE, and desktop, helping developers understand codebases, plan changes, and use multiple LLM providers.
openfang.sh
OpenFang is an open-source agent operating system built in Rust for running, configuring, and supervising autonomous agents, with local or self-hosted deployment, templates, custom manifests, messaging channels, and browser or API control.
continue.dev
Continue is an open-source coding agent for developers, acquired by Cursor, with its open-source codebase remaining available as a foundation for others.
kilocode.ai
Open source AI coding agent for VS Code, JetBrains, CLI, and cloud workflows, with model routing, BYOK, local models, and KiloClaw.
cline.bot
Cline is an open-source AI coding agent for VS Code, CLI, and enterprise, enabling code creation, editing, and verification through permission-based actions and usage pricing.
