Tek — a local-first AI agent for your files. Indexes your folders on-device, semantic + exact-keyword search across everything, chat with your files with cited sources, and safe file actions (dedupe, organize, AI rename) behind a preview you approve. Everything runs on your machine; nothing leaves it.
Highlights
- Hybrid retrieval: embeddings (bge-small, ONNX) + BM25 full-text fused with reciprocal-rank fusion, re-ordered by a local cross-encoder reranker. 19/19 eval probes top-1 on Windows, macOS, and Linux; honest empty results when nothing matches.
- Multi-turn chat grounded in your files via local LLM (Ollama), with clickable citations and follow-up questions.
- Zero prerequisites: every installer bundles a standalone Python 3.12 runtime — no system Python needed. First launch sets up the engine once (~1 min) and downloads the embedding models (~210MB total).
- Live re-indexing via file watcher; PDF and DOCX supported; deletes always go to the recycle bin.
Install
| Platform | File | Notes |
|---|---|---|
| Windows 10/11 (x64) | Tek.Setup.0.2.0.exe |
SmartScreen may warn (unsigned build) — More info → Run anyway |
| macOS (Apple Silicon) | Tek-0.2.0-arm64.dmg |
Unsigned: right-click the app → Open the first time |
| Linux (x64) | Tek-0.2.0.AppImage |
chmod +x then run |
All three builds passed the same CI gate: full retrieval eval + a GUI smoke test driving the real app.
Optional, for synthesized chat answers: install Ollama and ollama pull llama3.2:3b (search and passage answers work without it).