Thalius

Retrieval for your agent that scores 92% where Azure AI Search scores 77%

The retrieval layer you stop tuning. More answers right than Azure AI Search, Vertex AI Search and OpenAI File Search, in three endpoints or one MCP server. Your agent gets the passage and the page, or a refusal.

POST /search

Hippocampus

POST /search query: How does a customer reset their password?

passage: Open Settings, then Account, and choose Reset password. Help centre guide / Account / p. 4

passage: The customer gets a reset link by email. Help centre guide / Account / p. 4

POST /search query: What is on next quarter's roadmap?

refusal: the documents do not cover this

Sources
Help centre guidep. 4

What it does

Three steps to an answer with its page

  1. Drawing of documents being added

    Add your documents

    Upload files or connect a source. Hippocampus turns each document into a tree of sections with summaries.

  2. Drawing of a question typed into a computer window

    Your agent asks over API or MCP

    Call POST /ask or POST /search, or connect your agent to the MCP server.

  3. Drawing of an answer with its source page

    Get the answer with its page

    The passage, the document and the page, or a refusal if no passage supports an answer.

Hippocampus refuses the question if your documents hold no answer.

The problem

What goes wrong today

Drawing of many PDF files in a search result
01

Wrong chunk at scale

Your RAG demo worked on ten documents. At ten thousand it returns the wrong chunk.

Drawing of an old printed routine on a wall
02

A month of tuning

You spent a month on chunk sizes and rerankers and the eval score stayed where it was.

Drawing of an answer pointing to a page
03

Paying for whole documents

Your agent reads whole documents to find one paragraph, and you pay for each token.

What you can count on

Drawing of your own documents

More answers right

On FinanceBench, Hippocampus scores higher than Azure AI Search, Vertex AI Search and OpenAI File Search.

Drawing of an answer with its source page

Returns the page

Each passage comes with its document, section path and page range.

Drawing of an answer that says it does not know

Says so when it does not know

If no passage supports an answer, you get a refusal. A refusal is your signal to hand over to a person.

Drawing of a routine updated in one place

Works over API and MCP

Three endpoints, or one MCP server for coding agents and chat assistants.

For your team

People and agents ask the same documents

People ask in the web workspace. Your agents ask over MCP. Your developers call the API. You set access by role or function.

Drawing of one question sweeping across a whole collection of documents

Benchmark

Accuracy at cost

Basic vector search is cheap per query and misses more answers. Hippocampus costs more per query than that and gets more answers right.

Speed is the trade. Median search time is 1.8 s. Azure AI Search answers in 0.4 s and Vertex AI Search in 0.3 s, with lower scores.

  • 92.2% on FinanceBench, 150 analyst questions on 84 US company filings
    Hippocampus 92.2%
    Azure AI Search 76.6%
    OpenAI File Search 52.3%
    Vertex AI Search 40.6%

FinanceBench: 150 analyst questions on 84 US company filings. Hippocampus 92.2%, Azure AI Search 76.6%, OpenAI File Search 52.3%, Vertex AI Search 40.6%. GPT-5.4 mini wrote every answer.

Where it sits in your stack

Three endpoints or one MCP server

Your agent asks. Hippocampus answers from your documents, with the page, or says it cannot.

Your agent

Hippocampus

the memory

Your documents

POST /ask

An answer with its sources, or a refusal

A refusal is a signal the agent can act on: ask the user, hand over to a person or stop.

Select a part of the diagram to see its flow.

Under the hood

Hippocampus turns each document into a tree of sections with summaries. A question runs two searches over that tree: top down through the summaries, and bottom up with keywords and embeddings on the leaf sections.

Pick a question

How much notice do we need to give?

Model

writes the answer

Three months, in writing.

Sent to the registered address.

  • Supply agreement A three-year supply agreement.
    • Scope What is supplied, and when.
      • Delivery Delivery dates and places.
        • p. 6 Goods are delivered within 30 days.
        • p. 7
    • Termination How either side can end it.
      • Exit Handing back data.
        • p. 21
      • Notice How much notice to give.
        • p. 22 Three months' notice in writing. sent to the model
        • p. 23 Notice goes to the registered address. sent to the model
        • p. 24

Click a node to read its summary.

2 of 6 passages went to the model. The document did not.

Hippocampus merges the two rankings. Retrieval uses no generative model. The model you choose writes the answer from the passages.

Example with a fictional document.

Fits your setup

Models fail as volume grows. Hippocampus makes it scalable.

Ask a model about a few pages and it does well. Ask it across thousands and its answers get worse. Hippocampus keeps the model on the right pages, so it stays accurate as the collection grows.

Your sources

Connects to SharePoint, Google Drive and Box.

Your files

PDF, Word, PowerPoint, Excel and scanned pages.

Your hosting

EU-hosted by default, or in your own cloud or on your own servers.

Your access

You set who sees what by role or function.

Hippocampus One memory of your documents

Your model

You choose the model per workspace and can change it without importing again.

Pick another model. The memory stays as it is.

Questions

Frequently asked

What does it cost?

The Free plan holds 2,000 pages, needs no card and includes the API and the MCP server. The Starter plan is EUR 30 a month for 25,000 pages. The Corporate plan has custom pricing.

How fast is it?

Median search time is 1.8 s.

Can I get the passages without an answer?

Yes. POST /search returns passages only.

What happens when nothing matches?

Hippocampus refuses the question. A refusal costs the same as an answer.

Can I take my data out?

Yes. You export each document as a Markdown folder in Open Knowledge Format.

Start building. The Free plan includes the API and the MCP server.