← Overview · Compare

Polymathy vs Algolia + LLM

Hosted search-as-a-service plus a downstream LLM call (the "Algolia AI Search" pattern)

Algolia + LLM is the SaaS path to an answer engine. Polymathy is the self-hosted, public-web path. The dividing line is who owns the corpus: Algolia owns yours, SearxNG points at the public web.

Side by side

Grounded in the public docs and source on both sides — if a row reads as opinion, it is one we are willing to defend.

Dimension Polymathy Algolia + LLM Better fit
Hosting model Self-hosted Rust binary; you run SearxNG + the processor Hosted SaaS; Algolia indexes your corpus on their infrastructure Comparable
Corpus assumption Public web via SearxNG (or any metasearch endpoint) Closed corpus you push to Algolia as records Comparable
Retrieval signal Whatever SearxNG returns from upstream engines Tuned hybrid ranking with typo tolerance, faceting, personalisation Algolia + LLM
Chunking + embedding Delegated to a content-processor service of your choice AI Search ships embeddings + reranker as part of the product Algolia + LLM
Answer generation Out of scope — Polymathy returns chunks; you call the LLM Out of scope — Algolia returns hits; you call the LLM Comparable
Per-query cost Your VM + your processor + your downstream LLM Algolia per-search pricing + your downstream LLM Polymathy
Data residency On your hardware; corpus never leaves your network Records replicated to Algolia regions Polymathy
Admin UX None — operate with kubectl/systemd and the OpenAPI spec Full dashboard: index config, analytics, A/B tests, personalisation Algolia + LLM
Citation contract chunk_id → (source_url, text) enforced by response shape You stitch hit URLs into your prompt yourself Polymathy
Licensing GPL-3.0 source Commercial SaaS; free tier + paid plans Comparable

Pick Polymathy when

  • Your corpus is the public web, accessed through SearxNG, rather than a closed product catalogue or knowledge base
  • You are unwilling to ship the queries and the content to a third-party SaaS for indexing
  • You want the chunk-to-URL mapping to be enforced by a response shape, not by a downstream prompt template
  • You want one Rust binary you can run on a small VM, with no per-search pricing

Pick Algolia + LLM when

  • Your corpus is a closed set of records you control (a product catalogue, a help center, a docs site) and you want best-in-class hosted ranking
  • You want managed infrastructure with a UI library, dashboard, A/B testing, and SLAs
  • You want per-tenant analytics, click-through tracking, and personalisation built in
  • You are happy with per-search pricing and shipping your corpus to a SaaS index

Algolia + LLM is the right answer when you own the corpus and want hosted infrastructure with a dashboard. Polymathy is the right answer when the corpus is the public web and you want a single Rust binary you can read end to end. They do not really compete; they answer different questions about where the index lives.

Read the docs Read the notes