AI-powered research agent for journalists and newsrooms.
Search documents, ask questions, get answers with citations. Use cloud AI for convenience or run everything locally for maximum privacy.
- Ask questions, get answers — AI agent searches your documents and synthesizes answers with citations
- Your choice of AI — use cloud models (Anthropic, OpenAI) or run local models on-device
- Documents stay local — your files are processed on your machine, never uploaded
- Share with colleagues — sync collections peer-to-peer, no server required
- Work offline — with local models, everything runs without internet
You: "What do these documents say about the 2019 contract?"
↓
AI agent searches your documents
↓
Reads relevant files, synthesizes answer
↓
"According to the March 2019 filing [doc: contract-v2.pdf]..."
Your documents are indexed locally using milli for hybrid full-text and vector search. The AI agent—powered by cloud models or a local LLM via mistralrs—searches and reads documents to answer your questions.
Collections sync peer-to-peer using iroh, including pre-computed embeddings so collaborators don't redo the work.
This is a research project exploring:
- AI-assisted document research for journalism
- Practical P2P sync for document workflows
- Agent-driven search with hybrid full-text and vector retrieval
- Trust models for journalist collaboration
Not production-ready. Built to learn and prototype.
| Layer | Choice | Why |
|---|---|---|
| App | Tauri + Rust | Single binary, cross-platform, no Electron |
| Frontend | Svelte 5 | Fast, minimal |
| LLM | mistralrs | Local inference option (GGUF models) |
| P2P | iroh | Modern QUIC-based, handles NAT traversal |
| Search | milli | Full-text + vector search, runs locally |
| lopdf | Text extraction (opens in system PDF viewer) |
pnpm install
pnpm tauri dev # Desktop app with hot reload
pnpm tauri build # Release build (CPU)GPU-accelerated builds (faster inference):
# NVIDIA (requires CUDA toolkit)
pnpm tauri build -- --features cuda
# Apple Silicon
pnpm tauri build -- --features metal- Journalists managing document collections for investigations
- Newsrooms wanting AI-assisted research with privacy options
- Investigators who need to analyze and search large document sets
- Researchers exploring AI tools for document-based work
A research exploration of AI-assisted journalism tools.