RepoRecall AI

RepoRecall AI is a RAG-powered engineering assistant that searches a GitHub repository's issues, pull requests, commits, and code history to find similar past bugs and explain how they were fixed using traceable source evidence.

The project is built with Python, FastAPI, LangGraph, PostgreSQL, and Qdrant. It uses a modular-monolith architecture with a separate background worker for long-running repository indexing.

Current milestone

The current version can:

• register public GitHub repositories;
• create asynchronous repository-sync jobs;
• fetch issues, pull requests, and commits through the GitHub REST API;
• normalize and upsert the retrieved records into PostgreSQL;
• track indexing progress and failures;
• index and search engineering history with Qdrant;
• combine semantic and keyword retrieval, rerank results, and diversify sources;
• run a bounded LangGraph investigation with evidence-grounded citations;
• generate answers with OpenAI or a free local Ollama model.

Run locally

cp .env.example .env
# Add a GitHub token to .env to receive a higher API rate limit.
docker compose up --build

Open Swagger at http://localhost:8000/docs.

Free local answer generation

Install Ollama on the host, then download the default lightweight model:

ollama pull qwen2.5:3b

Configure the Docker API to reach Ollama in .env:

OLLAMA_BASE_URL=http://host.docker.internal:11434
OLLAMA_MODEL=qwen2.5:3b

No OpenAI key is required. Leave OPENAI_API_KEY empty; when both providers are configured, OpenAI takes precedence.

Basic flow

1. Register a repository with POST /api/v1/repositories.
2. Start ingestion with POST /api/v1/repositories/{repository_id}/sync.
3. Check progress with GET /api/v1/indexing-jobs/{job_id}.
4. Inspect collected data with GET /api/v1/repositories/{repository_id}/documents.

Architecture

Client
  |
FastAPI API ───────── PostgreSQL
  |                       |
  └── creates job         └── repositories, jobs, raw documents
          |
     Python worker ───── GitHub REST API

Qdrant stores document embeddings used by semantic and hybrid retrieval.