Every answer. Every source. One search.

The right answer, not
the most recent one.

Federated search finds documents. SWIRL 5 goes further: it lets your organization decide which result is authoritative. Teams collaboratively pin the official answer to any query - the approved clause, the governing policy, the correct precedent.

Every subsequent search - from a user, an AI assistant, or any system calling SWIRL via MCP - returns the endorsed result, not the model's best guess.

• Collaborative endorsement - approve, reject, or pin any result
• Canonical answers served to any AI calling SWIRL via MCP
• Version control - right document surfaces, outdated ones don't
• AI Yield Calculator - measure how often AI returns endorsed knowledge

Higher recall.
No vector database.

Meta's XetHub team benchmarked three retrieval models against a corpus of natural-language questions like "What is the capital city of Assyria?". Pure BM25 keyword search came last. Pure vector search did better. The winner was hybrid retrieval - pull the top 50 with keyword search, then re-rank with vector embeddings. That gold line is exactly how SWIRL works.

• BM25 alone misses synonyms - "capitals" won't match "capital city"
• Vector alone can return results that don't even contain the user's query - eroding trust
• Hybrid re-rank takes every exact match first, then falls back on vectors when it runs out

"You do not need document embeddings for RAG. Consider simpler information retrieval methods. Vector embeddings are still useful and are a great way to add a higher precision filter. No vector database necessary."

- XetHub research, 2024

SWIRL is cheaper too: no data movement, no embedding pipeline, no token spend on vectorization, and no risk of identical answers to queries the user perceives as different - a particular problem in legal and other complex domains.

Source: "You don't need a vector database" - XetHub, 2024 (archived)