AI-powered research agent for journalists and newsrooms.
Search documents, ask questions, get answers with citations. Your documents stay local; chat runs on your configured AI provider.
- Ask questions, get answers — AI agent searches your documents and synthesizes answers with citations
- Documents stay local — your files are stored, indexed, and searched on your machine, never uploaded
- Choice of provider — 25+ LLM providers for chat (OpenAI, Anthropic, Gemini, Groq, local servers like Ollama), configured with your own API key
- Share with colleagues — sync collections peer-to-peer, no server required
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 full-text keyword search (BM25). The AI agent—running on the rig framework with your configured provider—searches and reads documents to answer your questions. Only your conversation (questions and retrieved excerpts) goes to the provider; the documents themselves never leave your machine.
Collections sync peer-to-peer using iroh, so collaborators get full copies of documents and extracted text. Each peer builds its own search index locally.
This is a research project exploring:
- AI-assisted document research for journalism
- Practical P2P sync for document workflows
- Agent-driven search with full-text 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 | rig (25+ providers) | Remote inference via API (or local servers), user-configured |
| P2P | iroh | Modern QUIC-based, handles NAT traversal |
| Search | milli | Full-text keyword search (BM25), local |
| lopdf | Digital text extraction |
pnpm install
pnpm tauri dev # Desktop app with hot reload
pnpm tauri build # Release build- Journalists managing document collections for investigations
- Newsrooms wanting AI-assisted research without uploading documents
- 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.