Research moves faster when evidence is easy to trust.

Arxie is an AI research assistant that reads real papers (Semantic Scholar + arXiv), reasons across sources, and writes citation-grounded outputs you can verify.

• Current: v0.1.0 (released)
• Next: v0.2.0 (in planning)

• Live paper retrieval and citation-grounded Q&A
• Deep search (multi-hop paper analysis)
• Full-text PDF parsing for methods/results-level reasoning
• Literature review generation (ra lit-review)
• Citation influence tracing (ra trace)
• Confidence annotations (supporting vs contradicting evidence)
• Conversational mode (ra chat)
• FastAPI + Docker support

• Dashboard-based proposal workspace (not terminal-first)
• Iterative research proposal co-creation workflow
• Visual artifacts (mindmap, evidence map, logical tree, method pipeline, outcome matrix)
• Cross-artifact sync when users revise hypotheses/assumptions

(See docs/PRE-PRD-v0.2.md for discussion draft.)

Most assistants stop at summaries. Arxie is designed for researchers who need a defensible reasoning trail:

• read full papers, not just abstracts
• compare methods and contradictions across papers
• keep citations tied to claims
• show confidence based on evidence landscape

git clone https://github.com/mmTheBest/arxie.git cd arxie python -m venv .venv source .venv/bin/activate pip install -e . export OPENAI_API_KEY="sk-..."

# Ask a question ra query "What are recent approaches to long-context LLMs?" # Deeper multi-hop analysis ra query --deep "Compare LoRA vs QLoRA methodologies" # Literature review draft ra lit-review "attention mechanisms in computer vision" # Citation timeline ra trace "Attention Is All You Need" # Interactive session ra chat

uvicorn ra.api.app:app --host 0.0.0.0 --port 8000

curl -X POST http://localhost:8000/api/query \ -H "Content-Type: application/json" \ -d '{"query":"What are retrieval-augmented generation trade-offs?"}'

docker build -t arxie . docker run -e OPENAI_API_KEY="sk-..." arxie ra query "Your question here"

Using Arxie’s internal 100-question benchmark with GPT-4o-mini:

These are reported benchmark results, not a user quick-start workflow.

src/ra/ ├── agents/ # research, lit-review, chat behaviors ├── api/ # FastAPI app + request models ├── citation/ # citation formatting + confidence scoring ├── parsing/ # PDF parsing ├── retrieval/ # Semantic Scholar + arXiv + cache ├── tools/ # tool interfaces for the agent loop └── utils/ # config, logging, rate limiting 

MIT