Interactive architectural maps of your repo, show branches and commit diffs. AI

RepoMap, a new open-source tool, extracts repository structure without sending source code to an LLM, generating interactive architectural maps for human and AI agent use. The tool reduces token consumption by providing a deterministic structural representation that agents can reuse for reasoning and visualization. It integrates with coding agents like OpenCode and Claude to streamline architectural understanding.

Give coding agents architectural awareness. RepoMap extracts the structure of any repository without sending source code to an LLM, then generates an interactive architectural map that both humans and AI agents can understand. Modern coding agents spend a significant amount of time reconstructing a project's architecture. They repeatedly open files, follow imports, inspect folders, and consume thousands of tokens just to answer questions like: Where is authentication implemented? Which modules depend on the database? What architectural pattern does this project follow? How are components connected? RepoMap changes that workflow. Instead of asking the LLM to rediscover the architecture every session, RepoMap builds a deterministic structural representation once and lets the agent reason over that representation. significantly fewer tokens spent on repository understanding faster architectural reasoning interactive visualization for humans reusable architecture data for AI agents Features ✅ Deterministic repository analysis ✅ Zero source code sent to the LLM ✅ Interactive architecture graph ✅ Automatic architectural role detection ✅ Architecture pattern recognition ✅ Multiple graph layouts ✅ Editable graphs with persistent branches ✅ Git history visualization ✅ Large repository support RepoMap separates structure extraction from architectural reasoning. Repository │ ▼ Deterministic Analyzer │ ▼ RawAnalysis no source code │ ▼ LLM architecture reasoning │ ▼ RepoGraph │ ▼ Interactive visualization The analyzer never calls an LLM. It scans the repository and extracts: directory hierarchy imports function signatures module relationships git information The result is a compact RawAnalysis that describes the repository structure without exposing the source code. Instead of reading thousands of files, the LLM receives only the structured analysis. From that information it can: assign architectural roles identify architectural patterns improve module labels generate a visualization layout Only one reasoning step is required. RepoMap renders the generated architecture as an editable graph. Features include: React Flow visualization multiple layouts branch-based editing node inspection viewport culling persistent local storage Git-aware visualization RepoMap lets you explore repositories together with their Git history. inspect commits browse branches highlight added files highlight modified files highlight deleted files Future versions will expand this into full architectural diff visualization. Because repository exploration is mostly deterministic. Finding imports, discovering modules, following folders and extracting definitions does not require intelligence. LLMs should spend their context reasoning about architecture—not reconstructing information that software can extract automatically. This dramatically reduces token usage while producing a richer architectural representation. Designed for coding agents RepoMap was built to integrate naturally with tools such as OpenCode and Claude. Instead of repeatedly exploring the repository, agents receive a compact structural model that can be reused for reasoning, visualization and future analysis. git clone <repo-url cd repomap-pipeline-v2 npm install The skill is auto-discovered by OpenCode and Claude from .opencode/skills/repomap/ . Quick Start Analyze a local repository and serve the interactive map node .opencode/skills/repomap/cli.js open my-repo Or from code: import { analyze } from './.opencode/skills/repomap/index.js' const raw = await analyze { localPath: '/path/to/repo' } List all saved maps node .opencode/skills/repomap/cli.js list { "meta": { "repoName": "owner/repo", "estimatedSize": "small|medium|large", "languages": "TypeScript", "Python" , "totalFiles": 42, "totalModules": 8 }, "nodes": { "id": "layer src", "label": "Src", "type": "layer", "parentId": null, "files": , "metadata": {} }, { "id": "module src core", "label": "Core", "type": "module", "parentId": "layer src", "files": "core/index.ts" , "metadata": {} }, { "id": "file core index ts", "label": "index.ts", "type": "file", "parentId": "module src core", "files": "core/index.ts" , "metadata": { "language": "TypeScript", "lineCount": 150, "complexity": "medium" } } , "edges": { "id": "edge file1 file2", "source": "file core index ts", "target": "file utils helpers ts", "edgeType": "engineering", "strength": 3, "label": "imports", "confidence": "high" }, { "id": "edge layer module", "source": "layer src", "target": "module src core", "edgeType": "architecture", "strength": 5, "label": "contains", "confidence": "high" } , "overlay": { "version": 0, "nodeOverrides": {}, "edgeOverrides": {}, "manualNodes": , "manualEdges": } } node cli.js list List all saved maps node cli.js open <name Open a map by name, file name, or partial match All commands run from .opencode/skills/repomap/ . .opencode/skills/repomap/ analyzer.js Deterministic scanner: file tree, imports, definitions, signatures, git data index.js Orchestration: analysis → persist → serve visualization cli.js CLI wrapper list, open visual src/ React Flow-based interactive graph renderer published as @frannn2114/repomap-visual The pipeline is: analyzer.js extracts structure without calling an LLM index.js saves the result as a RepoGraph JSON and spawns a visual server cli.js provides list and open commands for saved maps MIT&und