🧠 CodeRecall

Context-aware local AI assistant for developers using Ollama, ChromaDB, and VS Code’s Continue extension.

CodeRecall ingests your entire codebase — including Git history and diffs — into a local vector database (ChromaDB), enabling RAG-augmented queries via Ollama models right inside VS Code.

No cloud APIs. No latency. Full control.

🚀 Features

🔍 Semantic code search across multiple languages
📜 Git commit + diff embedding for code evolution awareness
🤖 RAG integration with local Ollama models (e.g. LLaMA 3)
💻 VS Code Continue extension support
⚙️ Configurable with a simple config.ini

🧩 Project Structure

CodeRecall/
├── chroma_ingest.py           # Ingest codebase + Git into ChromaDB
├── chroma_context_provider.py # VS Code Continue context provider
├── config.ini                 # Ollama + Chroma settings
├── chroma-db/                 # ChromaDB persistence directory
└── config.json                # Continue extension config

🔧 Setup

1. Install dependencies

pip install chromadb requests

Make sure you have:

🦙 Ollama installed and running
✅ Continue Extension for VS Code
🐙 Git repo initialized (optional but recommended)

2. Configure `config.ini`

[ollama]
url = http://localhost:11434

[chroma]
persist_directory = ./chroma-db

📥 Ingest your project

python chroma_ingest.py

This loads:

Source code in common languages
Markdown and text docs
Git commit messages and full diffs

🧠 Add as a VS Code Context Provider

`config.json` for Continue

{
  "models": [
    {
      "title": "LLaMA 3 (Ollama)",
      "provider": "ollama",
      "model": "llama3",
      "apiBase": "http://localhost:11434"
    }
  ],
  "contextProviders": [
    {
      "title": "ChromaDB Search",
      "provider": "custom",
      "path": "./chroma_context_provider.py"
    }
  ]
}

✨ Usage

Launch VS Code.
Open the Continue sidebar.
Set "ChromaDB Search" as your context provider.
Ask your model questions about your codebase, architecture, or commits.

Example prompt:

“How does the Git ingestion pipeline work?”

📌 Notes

Default embedding model is nomic-embed-text (via Ollama).
Change n_results in chroma_context_provider.py for broader/narrower context.
Works offline, no API keys required.

🧪 Roadmap Ideas

Add OpenAPI spec ingestion
Enable full-text search fallback
Support for multi-repo ingestion
Optional chunking for large diffs or files

🛡 License

Creative Commons Attribution-NonCommercial 4.0 International license.

2.9 KiB Raw Blame History Unescape Escape