wikis

Wikis

AI-powered documentation generator that turns any code repository into a browsable, searchable wiki with architecture diagrams, code explanations, and an AI Q&A assistant.

Features

• Instant wiki generation — point at any GitHub, GitLab, Bitbucket, or Azure DevOps repo and get a comprehensive wiki in minutes
• 14+ languages — tree-sitter parsing for Python, TypeScript, Java, Go, Rust, C++, Ruby, PHP, Kotlin, Swift, and more
• AI Q&A — ask questions about any repository and get source-cited answers
• Deep Research — multi-step agentic research engine for complex codebase questions with tool calls and planning
• Mermaid diagrams — auto-generated architecture, sequence, and flow diagrams
• MCP Server — use your wikis directly from Claude Code, Cursor, or Windsurf
• Multiple LLM providers — OpenAI, Anthropic, Google Gemini, Ollama (fully local), AWS Bedrock
• Self-hosted — runs entirely on your infrastructure with a single docker compose up
• Authentication — built-in username/password, GitHub OAuth, Google OAuth

Quick Start

One command to install and run:

curl -fsSL https://raw.githubusercontent.com/arozumenko/wikis/main/install.sh | bash

The installer will prompt for your LLM provider, generate JWT keys, and start everything via Docker Compose.

Or manually:

git clone https://github.com/arozumenko/wikis.git && cd wikis
cp .env.example .env    # Edit: set LLM_PROVIDER and LLM_API_KEY
docker compose up -d

Windows: The install script requires bash (WSL or Git Bash). Alternatively, clone the repo, copy .env.example to .env, fill in your LLM key, and run docker compose up -d.

Open http://localhost:3000 and log in: [email protected] / changeme123

Change the default password immediately after first login.

How It Works

Repository URL
     │
     ▼
┌─────────────┐    ┌──────────────┐    ┌───────────────┐
│  Clone repo  │───▶│ Parse code   │───▶│ Build indexes │
│  (git)       │    │ (tree-sitter)│    │ (FAISS + BM25)│
└─────────────┘    └──────────────┘    └───────┬───────┘
                                               │
                                               ▼
                                    ┌──────────────────┐
                                    │  Generate wiki    │
                                    │  (LangGraph agent)│
                                    └────────┬─────────┘
                                             │
                                    ┌────────▼─────────┐
                                    │  Browsable wiki   │
                                    │  + AI Q&A + MCP   │
                                    └──────────────────┘

Architecture

Service	Port	Description
Web App	3000	Next.js 15 — React SPA + Better-Auth (JWT, OAuth)
Backend	8000	FastAPI — wiki engine, Q&A, deep research, MCP server

Both services ship as Docker images on GitHub Container Registry:

docker pull ghcr.io/arozumenko/wikis/web:latest
docker pull ghcr.io/arozumenko/wikis/backend:latest

LLM Providers

Provider	Models	Embeddings
OpenAI	gpt-4o, gpt-4o-mini	text-embedding-3-large
Anthropic	claude-sonnet-4-6, claude-haiku-4-5	requires secondary provider
Google Gemini	gemini-2.5-pro, gemini-2.0-flash	text-embedding-004
Ollama	llama3.2, qwen3.5 (local)	nomic-embed-text
AWS Bedrock	Claude, Titan (enterprise)	titan-embed-text-v1
OpenAI-compatible	Any (Together, Groq, vLLM, LM Studio)	varies

Set LLM_PROVIDER and LLM_API_KEY in .env to switch.

MCP Integration

Connect your AI IDE to Wikis for codebase-aware answers directly in your editor:

{
  "mcpServers": {
    "wikis": {
      "url": "http://localhost:3000/mcp"
    }
  }
}

Works with Claude Code, Cursor, Windsurf, and any MCP-compatible client.

Agent Skill

The wikis skill lets any AI agent browse wikis, ask codebase questions, run deep research, and generate new wikis — without opening a browser. It ships with this repo and works in three ways.

1. Claude Code plugin marketplace

/plugin marketplace add arozumenko/wikis
/plugin install wikis@wikis

Claude Code auto-discovers the skills/wikis/ directory and wires everything up.

2. npx — any IDE (Claude Code, Cursor, Windsurf, GitHub Copilot)

Copies the skill directly into the IDE skill directories of your current project:

# Install to all detected IDEs
npx github:arozumenko/wikis init

# Install to Claude Code only
npx github:arozumenko/wikis init --target claude