Incremental, local-first docs, now with Verilog/RTL

Incremental, local-first documentation with real Verilog/RTL support

Point BadDocs at a GitHub repo. It detects each file's language, extracts structure, uses an LLM to write clear docs, and makes everything fully searchable — accurate documentation in seconds.

Try it free See how it works

baddocs generate --repo ./ternarycore

$ baddocs generate --repo ./ternarycore

🔍 Detecting languages... rtl/ (Verilog), sim/ (Python)
🧩 Extracting structure... module ternary_mac: 8 ports, no multiplier inferred
🤖 Synthesizing docs... [local model fleet]
✅ unit → folder → repo hierarchy: 10 files, 5 folder/project summaries
♻️ Re-run after 1 edit: 1 file re-documented, 14 reused, only its folder chain rebuilt
📄 Output: ./bd-docs/ (Markdown)
▌

Features

Built for Real Codebases

BadDocs handles the messy, undocumented, legacy code that other tools ignore — with speed and precision.

Structural Extraction

Structural extraction for supported languages (full module/port/parameter parsing for Verilog) plus LLM-written prose for the rest. The LLM understands control flow, dependencies, and intent — producing clean, human-readable documentation that actually makes sense.

Export to Markdown & HTML

Export your docs in clean Markdown or styled HTML. Integrate with existing wikis, static site generators, or host them standalone. Your docs, your way.

Local-First LLM Support

Runs entirely on your machine with llama.cpp or Ollama. No API keys needed. Your code never leaves your environment. Cloud fallback available when you want it.

Incremental Re-Analysis

A content-keyed Merkle tree tracks every file and folder. Only changed files are re-analyzed on subsequent runs. The unit→folder→repo hierarchy means a single edit rebuilds only the affected file and its ancestor chain — unchanged runs do zero LLM calls. Perfect for CI/CD pipelines and active development.

GitHub App Auth

Secure access to private repositories via GitHub App authentication. No personal access tokens to manage. Enterprise-grade permissions with fine-grained control. A push now auto-updates only the affected docs — no full re-runs.

Language Support

Verilog and SystemVerilog are fully supported with real structural extraction, Python is supported, and other languages are in progress.

RTL / Hardware Description

Verilog SystemVerilog

General Purpose

Python

How It Works

How BadDocs Works

BadDocs analyzes your codebase, extracting structure and generating documentation with the help of local LLMs. It's fast, efficient, and designed for real-world codebases.

Analyze

Point BadDocs at your codebase and let it analyze the structure and dependencies.

Generate

Generate documentation in Markdown or HTML format, ready for integration.

Integrate

Integrate the generated documentation into your existing workflows or host it standalone.