Codag

See how your AI code actually works.

Codag analyzes your code for LLM API calls and AI frameworks, then generates interactive workflow graphs — directly inside VSCode.

If you find Codag useful, please consider giving it a ⭐ — it helps others discover the project!

vercel/ai-chatbot

langchain-ai/langchain

trycua/cua

You're building an AI agent that chains 3 LLM calls across 5 files. A prompt change breaks something downstream. Which call? Which branch? You grep for openai.chat, open 8 tabs, and mentally trace the flow. Again.

Or you're onboarding onto someone's LangChain project — 20 files, tool calls inside tool calls, retry logic wrapping everything. The README says "it's straightforward." It's not.

Codag maps it out for you:

• Extracts the workflow — finds every LLM call, decision branch, and processing step across your entire codebase
• Visualizes it as a graph — interactive DAG with clickable nodes that link back to source code
• Updates in real-time — edit a file and watch the graph change instantly, no re-analysis needed

Built for AI engineers, agent builders, and anyone maintaining code that talks to LLMs — whether it's a single OpenAI call or a multi-agent LangGraph pipeline.

Point Codag at your files and it maps out the entire AI pipeline — LLM calls, branching logic, data transformations — without any configuration.

Edit your code and the graph updates instantly using tree-sitter parsing. Changed functions get a green highlight so you can see exactly what moved.

Every node links back to the exact function and line number. Click a node to open the side panel, click the source link to jump straight to the code.

Export your workflow graphs as high-resolution PNG images — the entire graph or individual workflows.

Graphs automatically match your VS Code theme — light or dark. No configuration needed.

git clone https://github.com/michaelzixizhou/codag.git
cd codag
cp backend/.env.example backend/.env
# Add your Gemini API key to backend/.env (free tier: https://aistudio.google.com/apikey)
make setup

With Docker (recommended):

docker compose up -d

Without Docker:

make run

Verify: curl http://localhost:52104/health

VS Code: Search "Codag" in Extensions, or install from the Marketplace.

Cursor: Build and install the .vsix manually:

cd frontend && npx @vscode/vsce package
cursor --install-extension codag-*.vsix

1. Cmd+Shift+P / Ctrl+Shift+P → "Codag: Open"
2. Select files containing LLM/AI code
3. Explore the graph

The extension automatically registers a bundled MCP server when activated. This gives coding agents access to your workflow graph — no extra setup required.

The config is written to .cursor/mcp.json (Cursor) or .mcp.json (Claude Code) in your workspace.

Analysis Pipeline:

1. Tree-sitter parses your code into ASTs across 10+ languages
2. Pattern matching detects LLM API calls and framework usage
3. Call graph extraction maps function relationships
4. Backend (Gemini 2.5 Flash) identifies workflow semantics — nodes, edges, decision points

Live Updates:

• File changes trigger incremental tree-sitter re-parsing
• AST diffs determine which functions changed
• Graph updates instantly without LLM round-trip

Rendering:

• ELK (Eclipse Layout Kernel) for orthogonal graph layout
• D3.js for interactive SVG rendering
• Per-file caching with content hashing — only changed files reanalyze

• Hosted backend (no self-hosting required)
• Diff view: compare workflows across git commits
• Support for more languages and frameworks

Have a feature request? Open an issue.

For contributors working on the extension itself:

cd frontend
npm install
npm run compile

Then press F5 in VS Code to launch the Extension Development Host.

Contributions welcome! See CONTRIBUTING.md for full development setup.

Questions or feedback? Reach out at michael@codag.ai

MIT