Introduction

A locally-running documentation oracle that indexes web docs and injects relevant snippets into your coding context.

What is Mem- Oracle?

Mem- Oracle is a tool that helps AI coding assistants like Claude Code access and use documentation more effectively. It crawls documentation websites, indexes their content locally, and automatically injects relevant documentation snippets into your coding conversations.

Key Features

Seed-first indexing: Index the seed page immediately, then continue background crawling
Local storage: SQLite metadata + disk-based vector store (no external dependencies)
Pluggable embeddings: Local TF-IDF fallback, or use OpenAI/Voyage/Cohere APIs
Claude Code plugin: Hook scripts that auto-inject relevant docs into prompts
OpenCode plugin: Full integration with OpenCode editor
Optional MCP server: Explicit tool calls for search/index operations

How It Works

Index Documentation: Point Mem- Oracle at a documentation website (e.g., https://nextjs.org/docs)
Automatic Crawling: The crawler fetches pages, extracts content, and creates searchable chunks
Vector Embeddings: Content is converted to vectors for semantic search
Context Injection: When you ask a question, relevant documentation is automatically injected into your prompt

Quick Example

CLI

# Index Next.js documentation
curl -X POST http://localhost:7432/index \
  -H "Content-Type: application/json" \
  -d '{"baseUrl": "https://nextjs.org", "seedSlug": "/docs/getting-started"}'

# Search for relevant documentation
curl -X POST http://localhost:7432/retrieve \
  -H "Content-Type: application/json" \
  -d '{"query": "how to use server components", "topK": 5}'

Next Steps

Getting Started - Quick setup guide
Installation - Detailed installation instructions
Configuration - Configure Mem- Oracle to your needs
API Reference - Complete API documentation