Lochless vs a vector database

A vector database returns a fixed handful of chunks that look similar to your words. Lochless reads the documents in scope, up to your read limit, and returns the ones it judges relevant, however many there are.

The short version

Vector database: top results by similarity. You pick how many, and anything past the cut is lost.
Lochless: every relevant result in what it reads, page by page. Paging is free.
Vector database: you run chunking code, an embedding pipeline, the database and often a reranker.
Lochless: send raw text with an id and a date, up to 1 million characters per document. One API key, three calls.

Finding every answer

CUAD is a public set of 510 legal contracts with 41 questions and human-marked answers. Lochless read every contract and found 85% of the answers. A vector setup with a leading embedding model and reranker, keeping its top 1,000 results, found 45%.

Offline replay of the engine, October 2026, not the live API. A live-API run is in progress. Most RAG setups pass far fewer than 1,000 results to the LLM, so this cut favours the vector setup.

Questions a vector database struggles with

"Every customer who asked for a refund this quarter."
"Which contracts auto-renew with less than 30 days' notice?"
"Support tickets that describe a security issue, even if they never say 'security'."
"What did we promise Acme across all our calls and emails?"

These need every match, not the ten closest, and they need meaning, not similar words.

What it costs

$0.005 per page indexed (3,000 characters) and $0.10 per 1M tokens your queries read. Every query has a read limit, $1 by default, and a free dry run shows the exact price before it runs. A $100 monthly spend cap is on by default.

When a vector database is still the better fit

Lochless reads text, so a query takes seconds: about 2.5 seconds once warm, about 8 seconds for a cold first search. If you need the few closest chunks in milliseconds, a vector database is built for that. Lochless also works only from what is in your text, not outside world knowledge.

Try it

Sign in with Google at console.lochless.io. Your workspace opens right away with $5 of free credit. No card needed. Then follow the quickstart.

No account yet? Point any MCP client at https://api.lochless.io/demo/mcp to try the same tools free on 20 public US founding documents.

Start free Read the quickstart