About CodeMD.dev

What is CodeMD.dev, in plain terms?

CodeMD.dev reads your repository, extracts the facts (structure, callgraph, file relationships, UI flows), and writes them into one file: CODE.md. Drop that file into your repo and every AI coding agent that reads it starts from the truth instead of guessing.

Generate my CODE.md Jump to setup

At a glance

The moving parts, in one table.

PartWhat it isWhat it gives the agent
CODE.md One generated markdown file describing your repo. A factual starting map instead of a blank page.
Callgraph Function and class call relationships, extracted by parsing source directly. Who calls what, without reading every file.
File graph Cross-file import and dependency relationships. Which files are safe to touch together.
UI graph Buttons, forms, links, and routes mapped to the code that handles them. A path from what the user clicks to the code that runs.
Dashboard Where you connect a repo, run analysis, and generate CODE.md. One place to create and refresh the bundle.
Search Ask a plain-English question about the repo from the dashboard. An answer grounded in the same extracted evidence, with citations.

Core concept

One idea, explained twice.

AI coding agents are only as good as the context they start with. Left alone, an agent has to rediscover your repository's structure every single session: scan files, open likely candidates, trace imports, and guess what matters.

CodeMD.dev does that discovery once, from the actual source code (not from an LLM's guess), and writes the result into CODE.md: architecture, callgraph, file graph, UI graph, TODOs, and known gaps. Commit it to your repo, reference it from AGENTS.md or CLAUDE.md, and every agent session after that starts with facts instead of exploration.

Why use it

Useful when AI-assisted work starts to feel slow, repetitive, or ungrounded.

  • ✓Cuts repeated repo exploration, so agents spend fewer turns and fewer tokens finding their footing.
  • ✓Reduces hallucinated file paths, function names, and architecture assumptions.
  • ✓Gives every agent — Codex, Claude, Cursor, Copilot — the same factual starting point.
  • ✓Speeds up onboarding for new developers reading the codebase for the first time.
  • ✓Surfaces TODOs, gaps, and highly-connected "risky" functions directly from evidence, not opinion.
  • ✓Refreshable: regenerate CODE.md any time the repo changes so context never goes stale.

Deployment diagram

How CODE.md gets built and lands back in your repository.

Stage 1 — get the source
Your repo GitHub URL, connected repo, or uploaded archive
→
CodeMD.dev dashboard Downloads source, runs stats and file counts
→
Language parsers Python, JavaScript/TypeScript, C#, SQL, Java, plus an HTML/UI extractor
Stage 2 — build CODE.md and send it back
Extracted artifacts Callgraph, file graph, UI graph, repo text and comments
→
CODE.md One factual markdown file, generated from the evidence above
→
Back to your repo Download or commit CODE.md to the repo root

Runtime diagram

What happens after CODE.md exists, in two everyday paths.

Path A — an AI coding agent starts a session
Agent session starts Codex, Claude, Cursor, or Copilot opens the repo
→
Reads CODE.md Referenced from AGENTS.md or CLAUDE.md as @CODE.md
→
Starts from facts Architecture, callgraph, and UI graph loaded before the first edit
→
Fewer wasted turns Less rediscovery, fewer hallucinated assumptions
Path B — you ask a question in the dashboard
You ask a question Plain English, from the Code Intelligence Search panel
→
Evidence lookup Matches against callgraph, file graph, UI graph, and repo text
→
Grounded answer Generated from what was actually found, not from a guess
→
You review the evidence Source files and graph paths cited alongside the answer

Setup

Five steps, no installation required to get your first CODE.md.

1
Open the dashboard Go to /dashboard.
2
Point it at your code Paste a public GitHub URL, connect GitHub to pick a repository, or upload a code archive — pick whichever one option fits.
3
Run the analysis CodeMD.dev downloads the source, builds the callgraph/file graph/UI graph, and generates CODE.md.
4
Add CODE.md to your repo Download it or commit it straight into the repository root.
5
Reference it from your agent instructions Add one line so every session loads it automatically: 
@CODE.md

Benefits

What actually changes once CODE.md is in the repo.

Save time Agents skip repeated repo discovery and get to the actual task faster.
Save money Fewer exploration turns means less of your LLM budget spent re-reading the same files.
Fewer hallucinations The evidence policy tells the model what's known and what's excluded, so it stops guessing.
Fewer tokens A compact, pre-computed summary replaces thousands of lines the agent would otherwise re-read.

Generate your first CODE.md

No training, no runtime access, and no LLM summaries required for the truth layer — just point CodeMD.dev at your repo.

Generate CODE.md now