Skip to content
BranchesTags

Repository files navigation

DevDocs AI — AI-Powered Developer Documentation Engine

Code that documents itself — and stays accurate as the codebase evolves.

Next.js TypeScript Prisma PostgreSQL Gemini

Overview

Outdated documentation wastes developer hours and causes incorrect API usage. The fix is not more documentation — it is documentation that stays accurate automatically.

DevDocs AI treats documentation as a living asset. It ingests a GitHub repository, parses the source code into an Abstract Syntax Tree, uses an LLM to generate Markdown docs for every function, class, and API, and then keeps those docs in sync as the code changes. When a signature changes, the affected docs are flagged by severity and an updated draft is proposed for human review. A built-in chat assistant answers questions grounded in the generated docs — no hallucinated answers.

Core Features

Feature What it does
Code Ingestion & Parsing Connects a GitHub repo via OAuth, fetches source files, and parses JS/TS (Babel AST) and Python to extract functions, classes, methods, interfaces, parameters, return types, and async/export flags.
Documentation Generation Generates comprehensive Markdown per code unit — purpose, parameters, return values, side effects, usage examples, and edge cases.
Change Detection Detects code changes via GitHub webhooks and by re-analyzing a repo and comparing AST signatures against stored versions.
Staleness Flagging Classifies affected docs by severity: BROKEN, OUTDATED, or REVIEW_REQUIRED. Removed code units are flagged BROKEN.
Update Drafting Drafts an updated doc version reflecting the code change, presented as a side-by-side diff for approve/reject. Every decision is versioned.
Documentation Chat (RAG) Semantic Q&A over the docs using pgvector similarity search, with a keyword fallback. Answers are grounded in the docs and cite their sources.
Documentation Health Insights A dashboard that summarizes coverage and staleness for a repo, with an AI-generated health briefing on what to prioritize.

How It Works

GitHub Repository
      │  (OAuth + Octokit)
      ▼
Ingestion ──► AST Parser ──► Code Units (functions, classes, APIs)
                                  │
                                  ▼
                         LLM Doc Generation ──► Markdown docs (versioned)
                                  │
                                  ▼
                         Embeddings (text-embedding-004) ──► pgvector
                                  │
        ┌─────────────────────────┼──────────────────────────┐
        ▼                         ▼                            ▼
  Documentation Portal      Chat (RAG)              Change & Staleness Detection
                                                    (webhooks + re-analysis)
                                                            │
                                                            ▼
                                                  Update Drafts + Diff Review

AI model fallback

To stay available under rate limits or quota exhaustion, documentation and chat generation run through a fallback chain (see src/lib/gemini.ts):

  1. Gemini 1.5 Flash — fast, low cost, primary model
  2. Gemini 1.5 Pro — higher accuracy, used if Flash fails
  3. NVIDIA NIM → meta/llama-3.3-70b-instruct — open-weight fallback if both Gemini calls fail

Embeddings are always generated with Google text-embedding-004 (768 dimensions) and stored in PostgreSQL via the pgvector extension.

Tech Stack

Layer Technology
Frontend Next.js 15 (App Router), React 19, Tailwind CSS
Backend Next.js API Routes, TypeScript
Auth NextAuth.js + GitHub OAuth
AI (generation) Gemini 1.5 Flash → Gemini 1.5 Pro → NVIDIA Llama 3.3 70B
Embeddings Google text-embedding-004
Code parsing @babel/parser, ts-morph, custom Python parser
ORM Prisma 5
Database PostgreSQL + pgvector
Monitoring GitHub Webhooks, @octokit/rest
Validation / logging Zod, Winston

Local Setup

Prerequisites

1. Clone & install

git clone https://github.com/Durgaprasad-Developer/Dev-Docs.git
cd Dev-Docs
npm install

2. Configure environment

Create a .env.local file in the project root:

# Database
DATABASE_URL="postgresql://user:password@host:5432/dbname?schema=public"
DIRECT_URL="postgresql://user:password@host:5432/dbname?schema=public"

# AI
GEMINI_API_KEY="your_gemini_api_key"
NVIDIA_API_KEY="nvapi-your_nvidia_key"   # optional fallback

# Auth
NEXTAUTH_URL="http://localhost:3000"
NEXTAUTH_SECRET="generate_with: openssl rand -base64 32"
GITHUB_CLIENT_ID="your_github_oauth_client_id"
GITHUB_CLIENT_SECRET="your_github_oauth_client_secret"

# Webhooks (optional)
GITHUB_WEBHOOK_SECRET="your_webhook_secret"

Generate a NEXTAUTH_SECRET with openssl rand -base64 32.

3. Database

npx prisma generate
npx prisma migrate dev --name init

4. Run

npm run dev

Open http://localhost:3000.

Using the App

  1. Sign in with GitHub.
  2. Add a repository — select one of your GitHub repos or paste a URL.
  3. Analyze — the engine fetches files, parses them, generates docs, and builds embeddings. Status goes PENDING → ANALYZING → READY.
  4. Documentation — browse the generated docs per code unit.
  5. Chat — ask questions about the codebase; answers are grounded in the docs and cite sources.
  6. Insights — view documentation coverage and a per-severity staleness breakdown, and generate an AI health briefing on what needs attention.
  7. Review & approve — when code changes, affected docs are flagged and an updated draft is proposed; review the diff and approve or reject.

Database Schema

Table Purpose
repositories Tracked GitHub repositories
files Source files per repository (with content hash)
code_units Parsed AST nodes — functions, classes, methods, interfaces
documentation Active generated docs per code unit (with status)
documentation_versions Version history for review and rollback
embeddings vector(768) embeddings for semantic search
chat_history Developer ↔ assistant conversation log

Project Structure

src/
├── app/
│   ├── dashboard/        # Repository management
│   ├── repositories/     # Repo detail + file views
│   ├── documentation/    # Documentation portal
│   ├── chat/             # RAG chat interface
│   ├── insights/         # Documentation health insights
│   ├── diff/             # Update-draft diff review
│   └── api/              # API route handlers
│       ├── repositories/ # add, list, fetch, analyze
│       ├── docs/         # get doc, approve/reject update
│       ├── chat/         # RAG chat endpoint
│       └── webhooks/github/  # push-event receiver
├── services/
│   ├── parser.ts         # AST parsing (JS/TS + Python)
│   ├── generator.ts      # LLM doc generation & updates
│   ├── staleness.ts      # change detection & severity classification
│   ├── embeddings.ts     # pgvector embedding + similarity search
│   ├── chat.ts           # RAG pipeline
│   └── diff.ts           # LCS-based text diff
├── lib/
│   ├── gemini.ts         # AI model fallback chain + embeddings
│   ├── github.ts         # Octokit integration
│   ├── auth.ts           # NextAuth configuration
│   └── prisma.ts         # Prisma client singleton
└── components/           # Shared UI (Sidebar, etc.)

API Reference

# Repositories
POST   /api/repositories              Add a repository
GET    /api/repositories              List repositories
GET    /api/repositories/:id          Fetch one repository (with stats & code units)
POST   /api/repositories/:id/analyze  Parse + generate docs + embeddings

# Documentation
GET    /api/docs/:id                  Get documentation for a code unit
PATCH  /api/docs/:id                  Approve or reject an update draft

# Chat
POST   /api/chat                      Ask the RAG assistant
GET    /api/chat?repositoryId=        Fetch chat history

# Webhooks
POST   /api/webhooks/github           GitHub push-event receiver

Scripts

npm run dev       # start dev server
npm run build     # prisma generate + production build
npm run start     # run production build
npm run lint      # lint

Built with Next.js, Prisma, PostgreSQL/pgvector, and Google Gemini.

About

dev-docs-orpin.vercel.app

Resources

Readme
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

Packages

Contributors

Languages