💡 Rediscover the “why” behind engineering decisions — across pull requests, internal docs, and issues.
DevRecall is a local-first AI memory assistant for engineering teams. It ingests your GitHub PRs, issues, and internal docs to let you semantically query project history and get context-rich answers — powered by local LLMs and vector search.
- 🔁 Ingest GitHub PRs and Issues on demand
- 📄 Upload internal docs (
.txt,.md,.docx) - 🧠 Semantic search across all sources (Docs, PRs, Issues)
- 🔍 Smart chunking of long text for high-relevance matching
- 🎛 Filter by source type and file format
- 🔐 Token-based GitHub access for private repositories
- ✅ Runs 100% locally — no external API costs or data leaks
- Frontend: Streamlit
- Embeddings: SentenceTransformers (
MiniLM) - Vector Store: ChromaDB (local)
- Chunking: LangChain
- Ingestion: GitHub API + local file parser
- LLM (Optional): Ollama +
llama3(for summarization)
Devrecall/
├── data_docs/ # Place .md, .txt, .docx files here
├── scripts/
│ ├── ingest_github.py
│ ├── ingest_markdown.py
│ └── ingest_github_issues.py
│ ├── ingest_devrecall.py
├── streamlit_app.py # Main Streamlit UI
├── requirements.txt └── README.md
- Scripts folder is optional now.
# Clone this repo
git clone https://github.com/your-username/Devrecall.git
cd Devrecall
# Setup environment (recommended: conda)
conda create -n Devrecall python=3.10 -y
conda activate Devrecall
# Install dependencies
pip install -r requirements.txt- Ingests GitHub PRs, issues, and local docs (.md, .txt, .docx)
- Embeds them using sentence-transformers (MiniLM)
- Stores them in ChromaDB vector database
- Enables semantic querying via Streamlit UI
ℹ️ All document and GitHub ingestion is handled inside the app — no need to run separate ingestion scripts manually.
# 1. Install required dependencies
pip install -r requirements.txt
# 2. (Optional) Start your local LLM
ollama run llama3
# 3. Launch the app
streamlit run streamlit_app.py- “Why did we migrate to PostgreSQL?”
- “What decision was made in issue #45?”
- “Where is the reasoning behind the auth flow rewrite?”
- “What were the alternatives to Firebase?”
- 🤖 LLM-powered summaries and Q&A using local Ollama
- 🧭 Interactive document explorer with full-chunk linking
- 🔄 GitHub PR timeline view with decision threads
- 🧩 Multi-user project mode with per-user storage
- 🌐 Qdrant migration for scalable vector search
- 📚 Exportable search insights and summaries (PDF, Markdown)
- Feel free to fork and customize!
- Ideas, issues, or pull requests are welcome.
- MIT License — free to use, modify, and share.
- Developed by Rishabh Balaiwar. Feel free to reach out for questions or collaboration opportunities!
- For support or inquiries, contact me at:
- Email: rbalaiwar@gmail.com
- GitHub: Your GitHub Profile