01
Agentic search, with citations
A multi-step tool-calling agent searches the web, Wikipedia, and your own corpus, then synthesizes a single cited answer. It decides which sources to reach for and when to stop.
Self-hostable AI search
AI search you can self-host
Perplexity-style search over the open web and your own files. Plug in your LLM key, run the whole stack yourself. This is a pattern and a working self-host example, not a hosted product.
$ rh search "how does RRF rank fusion work"web 6 results (tavily)corpus 3 chunks (pgvector, qwen3-1024)$ rh recall "our embeddings config" -k 30.91 infra/embeddings.md#L120.88 notes/ingest-pipeline.md#L40
$ rh --whoami
search Tavily, or self-hosted SearXNG
corpus your files, embedded in pgvector (qwen3, 1024-dim)
models bring your own key — any OpenAI-compatible gateway
cli rh — search, recall, research, ingest
deploy docker compose, one stack
license Apache-2.0
02
Your files, made searchable
Ingest PDFs, documents, audio, and video. The bundled job processor parses and transcribes them, embeds the chunks into a pgvector corpus, and the agent searches it alongside the open web.
03
Bring your own key
Point it at any OpenAI-compatible gateway and pay for your own inference. No subscription, no per-seat pricing, no vendor lock-in. The same key drives the web app and the CLI.
04
Self-host the whole thing
One docker-compose stack: Postgres with pgvector, object storage, the job processor, and the app. Run it on a laptop or a homelab box. Your data and queries stay on your infrastructure.
05
A CLI built to be scripted
rh exposes search, recall, research, and ingest from the terminal. It is a thin tool over your search backends and LLM gateway — easy to pipe, automate, and shell out to from other agents.
Bring it up
git clone https://github.com/protoLabsAI/rabbit-hole.io
cd rabbit-hole.io
cp .env.example .env # add your LLM key
docker compose upThen open the docs for the quickstart. Bring your own LLM key; everything runs on your machine.