MAVYN is a terminal-based research assistant. Drop your PDFs into a folder, run mavyn, and start asking questions in plain English. Everything runs locally — no uploads, no subscriptions.
- Q&A — ask anything about one or many papers
- Summarize — get structured summaries by section
- Compare — side-by-side comparison of methodologies, findings, etc.
- Literature review — auto-generate a full review from your library
- Semantic search — local FAISS + BM25 hybrid retrieval
- Similar papers — find related work in your library (+ optional arXiv)
pip install -r requirements.txtSet a free Groq API key (takes 30 seconds):
echo "GROQ_API_KEY=your_key_here" >> ~/.MAVYN/.env$ mavyn
╭─────────────────────────────────────────────╮
│ Welcome to MAVYN 🤖 │
│ │
│ Commands: /sync ~/Papers │ /list │ /help │
│ Example: tell me about paper 5 │
╰─────────────────────────────────────────────╯
MAVYN> /sync ~/Papers
# Scans, renames, and embeds all PDFs — one command
MAVYN> /list
# Shows your papers with IDs
MAVYN> tell me about the methodology in paper 3
MAVYN> compare papers 1 and 5
MAVYN> find papers similar to paper 2
MAVYN> /review # generate a literature review
| Command | What it does |
|---|---|
/sync [dir] |
Scan, rename, and embed all PDFs in a directory |
/sync --watch |
Watch for new papers automatically |
/list |
List all indexed papers with IDs |
/review |
Generate a full literature review |
/model |
Show LLM status and rate-limit cooldowns |
/help |
Full command reference |
/exit |
Quit |
Natural language is always on — just type your question without any prefix.
All papers and embeddings stay on your machine. The only outbound calls are:
- Groq / Gemini / OpenRouter — only for LLM answers, only if you set an API key
- arXiv — only if you use
--arxiv(sends a keyword query, never your PDFs)
No account required to use local features.
- Python 3.10+
- ~500 MB disk space (embedding model)
- API key optional — search and indexing work without one
MIT — see LICENSE