INTENT-AWARE BM25 RETRIEVAL

Answers Pulled From the Right Page

Not just any page that mentions the right words. My Chat Assistant recognizes what kind of question it's being asked and ranks your content with BM25, the industry-standard search algorithm behind Elasticsearch and Lucene, refined from the classic TF-IDF weighting scheme.

my chat assistant

Retrieval Engine Capabilities

Grounding answers in your own indexed content.

Understands the Intent Behind a Question

Recognizes what kind of question it's being asked, a price question, a how-do-I question, a comparison, and widens the search to the vocabulary that kind of content is normally written in. A visitor asking "how much does it cost?" finds your pricing page even if that page never uses the word "cost."

TF-IDF & BM25 Retrieval

Built on proven information-retrieval foundations: TF-IDF term weighting, refined by BM25, the same ranking function used by Elasticsearch, Lucene, and Solr, with stemming and synonym expansion for accurate matches.

Weighs Your Most Authoritative Content

A current pricing page, FAQ, or product spec is treated as a more authoritative source than a blog post or news article that happens to mention the same words in passing.

Draws From Multiple Sources

When several pages are each a little relevant, answers draw from all of them instead of quoting the same page three times over.

Retrieval-Only by Default

Out of the box, answers come directly from your indexed content. No generation, nothing invented, nothing added beyond what's actually on your pages.

Optional RAG, Strictly Grounded

Connect an AI provider (Anthropic, Google Gemini, or any OpenAI-compatible endpoint) and it composes a fluent answer, but only from the specific content BM25 already retrieved and ranked, never from the model's general knowledge. This is Retrieval-Augmented Generation (RAG): the credible alternative to a model that simply makes things up.

Automated Indexing Pipelines

Ingest Markdown, HTML, JSON, or plain text documentation directly into local SQLite tables, with zero external database costs.

Answer-First Citation Schema

Structure returned answers with direct reference links pointing to the exact documentation source page for complete auditability.