EU hosted
The default. Your documents are stored and processed in the EU, and they are not used to train models.
How it works
Add anything, Hippocampus finds the correct answer
Memory layer
A memory layer sits between your documents and the AI models your organisation uses. It reads your material once and organises it into a structured memory: the sources, summaries built across them, and an index over both. When someone asks a question, nothing searches everything again. The question goes to the memory, and the memory returns the exact material that answers it, with the source attached. One memory serves every person and every agent, so the organisation answers from the same knowledge rather than from a dozen private copies of it.
The problem
Why can't our AI answer from our own documents?
Your organisation already knows the answer, but it's spread across thousands of documents that today's AI tools can only read a few of at a time, so they answer from fragments and can't show you where the answer came from.
Does Hippocampus actually get the answers right?
- 100% Retrieval accuracy Across every corpus tested, the right material is retrieved for the question asked.
- 97% Answer accuracy The rest is a model limitation, not ours. Hippocampus isn't locked to one model.
- 0 Hallucinations When the material doesn't say it, the answer says so instead of guessing.
Use cases
What do teams use it for?
Decision support
Cited answers behind a decision
Leadership and analysts ask the question a decision rests on and get an answer that quotes the contract, report or policy it came from.
Internal knowledge
One memory instead of the colleague who remembers
The answer no longer depends on finding the one person who knows where to look. Everyone asks the same shared memory, and every answer names its source.
Assistants
A support chat that quotes instead of guessing
A support or copilot chat grounded in your own documents: it answers from the passage it found, and says so when your documents do not cover the question.
Agents
Grounded answers in the middle of a workflow
Automated workflows call the API or MCP mid-task and get a cited answer back, limited by the same permissions as a person.
Permissions
Who can see what?
Hippocampus has permission controls, allowing organisations to control what documents employees can view. Based on role, function or manual edits.
As the head of procurement: what notice period applies if we end the logistics supplier agreement?
90 days' written notice. Amendment 2, section 7.3, replaced the original 30 days.
amendment-2-2025.pdf, page 2
As a project coordinator: what notice period applies if we end the logistics supplier agreement?
The documents you can read do not include the supplier agreement. The purchasing policy says changes to supplier contracts go through procurement.
purchasing-policy-2025.pdf, page 3
Deployment and security
EU hosted by default, in your own cloud, or fully on-premises. The product is the same in all three, and your documents are not used to train models.
Your own cloud
The same product, run in your organisation's own cloud.
Fully on-premises
The same product, run fully on your own hardware.
Model-agnostic
Use the model your organisation already approved: Claude, GPT, Gemini or an open-weight model. Switch without re-ingesting anything.
Fits your stack
Does it work with the tools we already use?
Document sources
Hippocampus supports PDF, DOCX and MD files.
MCP
Agents connect over MCP, with streamable HTTP and OAuth, and get cited answers mid-task.
API
Ask questions and walk the document tree from your own code.
Workspace
A ready-to-use workspace: add your documents and start asking, with nothing to build or set up.
Customer stories
Who runs it today?
Eldercare, field operations live pilot
Care staff retrieve procedures on the phone, between tasks
Purpose: Ensure that care providers follow procedure and law at the point of care. Chat box in staff's phones.
Result: Accurate where their earlier AI tools failed. 50-90% fewer tokens per query at quick response times.
296
documents
Pharmaceutical research, regulatory affairs live pilot
Regulatory reviews and labels, thousands of pages, answered with the page cited
Purpose: Analyze and cross-reference existing external clinical trials to guide their R&D spending.
Result: 16 of 16 questions answered correctly. Traditional AI failed the same corpus.
10
documents (2,200 pages total)
Questions
Frequently asked
Does Hippocampus support single sign-on?
Yes, SAML and OIDC single sign-on are part of the Corporate package. Sign-in runs on Google Identity Platform with a separate tenant per workspace, and signed tokens are checked by the application and again by each backend service.
Can we export audit records?
Yes, audit export is part of the Corporate package. Inside the service every security-relevant operation on the document store goes to a numbered, hash-chained audit log, and each answer resolves to a byte range in a named source, so access can be traced to the passage.
Can we have a dedicated EU stack instead of the shared one?
Yes, on Corporate the service runs on a shared or a dedicated EU stack, chosen and set up with you; Free and Starter run on the shared EU stack. On any stack each workspace has its own search collections, and its identity is bound into every encryption, so one workspace's data cannot be decrypted as another's.
Do you sign a data processing agreement?
Yes, every customer is covered by our standard data processing agreement, accepted with the terms of service, and Corporate customers can negotiate a separate one. It meets Article 28 of the GDPR, makes the sub-processor list part of the agreement and commits us to breach notice within 72 hours. Read it on the data processing agreement page.
How does a demo work?
You bring a set of documents and the questions you want answered, and we run them live during the call. To book one, write to letstalk@thalius.ai.
Can we host Hippocampus ourselves?
Yes, Hippocampus can run in your own cloud, fully on-premises or in any other environment you control, as part of the Enterprise package. Own hosting is set up with us, the product is the same as on our EU cloud, and the API and MCP differ only in the base URL. With own hosting, every component including model inference can run inside your environment.
Can we choose the AI models or use our own model key?
Yes. Hippocampus works with any language model. You choose which model answers, set per workspace; on Corporate you can use your own model key or have administrators choose the models. Changing the answer model needs no re-import.
Book a demo
Bring a set of documents and the questions you want answered, and we will run them live during the call.