Turn weeks of literature review into hours. LitAI lets you have research conversations with your entire paper collection - ask questions across multiple papers and get cited, contextual answers. Whether finding your research direction or unblocking active experiments, LitAI synthesizes literature to accelerate discovery.
Note
Development discontinued due to contractual obligations. This experimental codebase prioritized speed of iteration over production-quality standards and is not representative of my professional engineering practices.
Literature reviews take weeks. You read dozens of papers, lose track of insights, and struggle to synthesize findings across documents. Existing tools help you find or store papers - but not understand them together.
LitAI is the only tool that lets you have research conversations with your entire paper collection:
- Discovery: Search millions of papers using natural language
- Collection: Save papers locally with automatic ArXiv PDF downloads
- Context Building: Add your notes, select which papers and sections to analyze
- Synthesis: Ask questions across multiple papers and get cited, contextual answers
This synthesis capability transforms how you work:
- Finding Your Research Question: Explore a field systematically, discover gaps, understand contradictions
- Active Research Support: Get immediate answers to operational questions that arise during experiments, debugging, or analysis
Unlike AI writing tools, LitAI helps you discover your research direction through literature understanding, not by choosing for you.
- Graduate Students: Navigate unfamiliar literature to find and refine research questions
- Active Researchers: Unblock experiments with immediate synthesis of relevant methods
- Engineers: Find academic solutions to technical problems in production
- Research Teams: Build shared understanding across collaborative projects
uv tool install litai-research && export OPENAI_API_KEY=sk-... && litai- Python 3.11 or higher
- OpenAI API key (Get one here)
Warning
Currently, papers can only be downloaded from ArXiv. Support for importing your own PDFs is coming soon via /import.
Get your API key from platform.openai.com/api-keys
Permanent setup (recommended):
macOS:
echo 'export OPENAI_API_KEY="sk-..."' >> ~/.zshrc && source ~/.zshrcLinux:
echo 'export OPENAI_API_KEY="sk-..."' >> ~/.bashrc && source ~/.bashrcCurrent session only:
export OPENAI_API_KEY=sk-...First install uv, then:
# Using uv (recommended)
uv tool install litai-research
# Alternative: using pipx
pipx install litai-researchTip
If litai command not found, restart your terminal.
# Get latest stable updates
uv tool upgrade litai-research
# Alternative: using pipx
pipx upgrade litai-researchlitaiTell LitAI about your research focus for better responses:
/promptThis opens an editor where you can describe your background, interests, and preferences. LitAI includes this in every conversation to tailor its responses.
The Workflow:
- Find: Search for papers → "find papers on transformers" or
/find transformers - Save: Add to collection → "add papers 1-3" or
/add 1-3 - Organize: Add notes/tags → "add a note" or
/note - Analyze: Build context → "add paper to context" or
/cadd <paper> - Synthesize: Ask questions → "what methods do they use?" or
/synthesize
Important
Only papers in your context are analyzed. Collection stores everything; context is your active analysis set.
Note
LitAI understands natural language - just chat with it. Want more control? Use /commands instead. Mix both freely.
Commands: For a complete list of commands, use /help in LitAI. For detailed information about any specific command, use <command> --help (e.g., /add --help).
AI Models: LitAI uses two models for optimal performance:
- Large model (GPT-5): Used for
/synthesisqueries - Small model (GPT-5-nano): Used for search, extraction, and simple operations
These can be customized in settings, but we recommend the defaults for best results.
Exploring a new field:
→ Find recent papers on vision transformers
→ Add the top 5 papers to my collection
→ Add ViT and DINO papers to context with abstracts
→ What are the main architectural innovations?
Debugging your implementation:
→ Find papers about transformer memory efficiency
→ Add papers 1-3 about flash attention
→ Add them to context with full text
→ How do they handle the quadratic complexity problem?
Finding research gaps:
→ Search for graph neural network survey papers
→ Save all the recent surveys
→ Add top 3 surveys to context
→ What problems do they identify as unsolved?
LitAI stores all data locally in ~/.litai/:
litai.db- SQLite database with paper metadata and extractionspdfs/- Downloaded PDF fileslogs/litai.log- Application logs for debuggingconfig.json- User configurationuser_prompt.txt- Personal research profile
The LitAI database (~/.litai/db/litai.db) is a standard SQLite database that you can explore and manage with any SQLite-compatible tool. We recommend Beekeeper Studio for its user-friendly interface, but you can use any database tool you prefer.
To open the database in Beekeeper Studio:
- Download and install Beekeeper Studio
- Open Beekeeper Studio and click "New Connection"
- Select "SQLite" as the database type
- Click "Browse" and navigate to:
~/.litai/db/litai.db- macOS tip: Hidden files (starting with
.) may not be visible in Finder by default. PressCommand + Shift + .to show hidden files
- macOS tip: Hidden files (starting with
- Click "Connect"
You can now browse tables, run queries, and explore your research data directly.
Semantic Scholar's public API can experience high load, leading to search failures. If you encounter frequent issues:
- Wait a few minutes and try again
- Consider requesting a free API key for higher rate limits: Semantic Scholar API Key Form
- Email issues to harmonsbhasin@gmail.com
- Logs for debugging:
~/.litai/logs/litai.log
Due to a non-compete, I can no longer work on this code.
# Using uv (recommended)
uv tool install --prerelease=allow litai-research
# Using pipx
pipx install --prerelease litai-research- Created by Harmon Bhasin and Alex Wilf
- Alex Wilf conceived the original concept and developed an initial prototype that established the foundation for this project.
- Harmon Bhasin rebuilt the entire codebase from scratch, incorporating feedback from Alex Wilf and introducing substantial enhancements to create the current feature-complete application.
- Powered by Semantic Scholar API and OpenAI API
This project is open source and available under the MIT License.