mavyn 2.3.0

Local-first paper manager with semantic search and LLM reasoning

Chat with your research papers — locally, privately, for free.

---

---

What it does

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)

---

Install

pip install -r requirements.txt

Set a free Groq API key (takes 30 seconds):

echo "GROQ_API_KEY=your_key_here" >> ~/.MAVYN/.env

---

Quick start

$ 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

---

Commands

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.

---

Privacy

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.

---

Requirements

• Python 3.10+
• ~500 MB disk space (embedding model)
• API key optional — search and indexing work without one

---

License

MIT — see LICENSE