Documentation & Help

Learn how to configure RAG, connect external agents via MCP, and optimize your PRAI workflow.

Retrieval-Augmented Generation (RAG)

How it works

PRAI indexes your entire codebase into a vector database (LanceDB) to provide deep context during security audits.

Local vs Gemini Embeddings

Local mode uses Transformers.js (free, private), while Gemini mode uses Google's API for higher precision.

Incremental Indexing

The system only re-indexes changed files, saving significant time and compute resources.

Model Context Protocol (MCP)

Connecting External Agents

Use the SSE endpoint (http://localhost:3000/mcp/sse) to link PRAI with Cursor, Claude, or other AI tools.

Available Tools

Exposes "list_indexed_repositories" and "search_codebase" to external agents.

Stdio Support

You can also add external MCP servers to PRAI in the Settings menu.

GitHub Integration

Webhook Setup

Configure GitHub to send PR events to /api/webhook for automated analysis.

Manual Submissions

Paste any PR or Commit URL into the "Manual URL" field on the dashboard for instant scanning.

Authentication

Requires a Personal Access Token (PAT) with "repo" scope to access code diffs.

Security & Optimization

API Hardening

Built-in protection against DoS attacks with request limiting and strict input validation.

Performance

Uses infinite scrolling and virtualized lists to handle thousands of analysis records smoothly.

SQL Aggregation

Statistics are calculated on the server using optimized SQL for instant dashboard updates.

Detailed Setup Guide

Comprehensive walkthrough for deployment and configuration.

VIEW DOCS

Best Practices

How to write secure PRs and optimize your analysis results.

READ MORE

Community & Support

Get help from the developers and other PRAI users.

GET SUPPORT

Quick Setup Checklist

Ensure your instance is running with peak performance and security.

Review Settings