Holusight indexes PDFs, Office files, markdown, and code on your infrastructure. Hybrid retrieval fuses keyword and semantic search; every result links to the exact source chunk.
Teams store policies, contracts, runbooks, and code across folders and wikis. Generic chatbots hallucinate. grep misses concepts. Holusight is built for retrieval you can verify — ranked chunks with file, page, or line citations.
SQLite FTS5 BM25 + LanceDB vectors merged with reciprocal rank fusion. No LLM required for search.
Optional LLM synthesis (Claude, Azure OpenAI, OpenAI, or Ollama). Retrieved chunks are always shown alongside the answer.
FastAPI server + browser UI, API key auth, read-only document mount, persistent index volume.
python -m codesight index|search|ask and the CodeSight class for automation.
See the citation contract for how file types map to page vs line metadata.
Holusight is a focused consulting product — not a Sourcegraph or Glean replacement.
| Good fit | Not yet |
|---|---|
| Single-team pilot over a document folder | Company-wide SSO |
| Cited search across PDFs + code | Live M365 / SharePoint sync |
| On-prem or customer-cloud VM | Multi-tenant SaaS on holusight.com |
| Hybrid BM25 + semantic retrieval | Graphify code-graph integration |
Indexing, embeddings (default model), BM25, vector search, and the index files on your disk or volume.
Only when you configure Ask with a cloud LLM, or opt into Voyage/OpenAI embeddings. Use Ollama for fully local answers.