whereis <my-command>

Natural language search across tldr-pages CLI documentation.

1. No hallucination on commands — every answer is grounded in tldr-pages documentation, so the commands are real and tested
2. Structured, consistent output — always a command, an explanation, and a source; no conversational filler
3. Ranked alternatives — 3–5 results ordered by commonality, from the standard answer to the edge cases
4. No context needed — one input, one output

Example queries:

• "recursively find files modified in the last 24 hours"
• "watch a log file and filter for errors"
• "list open ports on this machine"
• "copy files over SSH"
• "show disk usage sorted by size"

---

Status

Live at https://container-service-1.gqceswqwzkchr.us-west-2.cs.amazonlightsail.com

---

Stack

• Corpus: tldr-pages (~7,000 pages, MIT licensed)
• Embeddings: sentence-transformers all-MiniLM-L6-v2 (local, no API key required)
• Vector store: Chroma
• RAG: LangChain + Gemini 2.5 Flash
• Backend: FastAPI with per-IP rate limiting
• Frontend: Vanilla HTML/JS

---

Running locally

1. Ingest the corpus (one-time, re-run to update):

cd ingest
pip install -r requirements.txt
python ingest.py

2. Start the backend:

cd backend
pip install -r requirements.txt
GOOGLE_API_KEY=your-key uvicorn main:app --reload

Open http://localhost:8000.

---

License

MIT