FAQ
Frequently asked questions.
Straight answers about what Polymathy is, what it isn't, and how the pieces fit together.
Is Polymathy a RAG product?
No. Polymathy is the fetch-and-chunk seam of an answer engine — infrastructure for building one, not a turnkey product. It returns a chunk map; retrieval-augmented generation happens in the LLM call you make downstream.
What does Polymathy actually include?
An actix-web service exposing GET /v1/search, an apistos-generated OpenAPI spec, four API doc UIs (Swagger, ReDoc, RapiDoc, Scalar), the SearxNG client, the parallel processor fan-out, and the chunk-map assembly. That is the whole binary.
What does it not include?
No embedding model (the processor provides it), no LLM, no reranker, no citation renderer, no auth or rate limiting, and no persistent index. Those are intentionally out of scope.
What is the response shape?
A JSON object mapping sequential u64 chunk IDs to a two-element array: [source_url, text]. Every chunk carries its source URL, so citations are enforced by the shape rather than by a downstream prompt.
Do I have to run my own SearxNG?
You point SEARXNG_URL at a SearxNG instance with JSON output enabled. It can be your own, a private one, or any instance you are permitted to use. Polymathy never crawls or indexes on its own.
What is the content processor?
An HTTP service behind PROCESSOR_URL that takes a URL plus a chunking/embedding config and returns chunks with embeddings. Fetching, boilerplate stripping, chunking, and embedding all happen there — so you choose the chunker and the model.
Does it persist anything between requests?
Not in v0.2. The USearch index is created per request and discarded. This keeps the service stateless and horizontally scalable; nothing is cached across queries yet.
How many URLs does it fetch per query?
The top ten result URLs from SearxNG, dispatched to the processor in parallel.
What language and license is it?
Rust (actix-web 4, tokio, reqwest, usearch), released under GPL-3.0 and published as the polymathy crate on crates.io.
How does Polymathy differ from Onyx or Algolia + LLM?
Onyx is a full answer-engine platform with connectors and a chat UI; Algolia + LLM is hosted search over a corpus you own. Polymathy is a single self-hosted Rust service over the public web via SearxNG. See the comparisons for a row-by-row breakdown.
Still unsure whether Polymathy fits your stack? Get in touch or read the comparisons.