AI memory for your company
Hippocampus is an AI memory service with LLM chat, agentic memory via MCP and API and your smart company wiki. Get AI productivity with your current processes, compliance and security. Hippocampus turns contracts, procedures, filings and reports into a wiki your team, AI agent or LLM can read and question. Every answer quotes its source page. Built and hosted in the EU.
High-risk substances are checked by a second qualified nurse before they are given1. A large expense needs written approval from the finance director before it is incurred.2
What you get
Power your company with AI by enhancing your business processes, compliance and efficiency with your data and LLMs
Documentation
Plain documents, tables and guides from your company document storage with existing access rights, security schemes - become a company brain and wiki with a help of LLM models. Existing processes, super-powered by AI, without sensitive data disclosure - engage your teams for more comfort and efficiency.
AI innovation
Modern markets require permanent innovation adoption and business process improvement - knowledge factory with wiki does not break an existing business and is a helper to analyze, improve and benefit on each step. AI will serve your team, not the team to AI.
Sets out who may prepare, administer and document medication in the units, the double-check rule for high-risk substances, how delegation is recorded and signed, and how deviations are reported within the shift.
SectionsEfficiency
Smarter algorithms make your team, documents, finances and KPIs better. You are free to use your document management and any LLM with Hippocampus - it responds faster and more precise even with smaller and cheaper LLM models. Use it via browser, chat bot, API or in agent with MCP. Answers always point to the original sources.
Insulin may be administered by a nurse, or by care staff holding a written delegation for that resident 1. Before a dose, the second-person check and the delegation reference are signed on the medication list 2.
Free account
How it works
The expensive thinking happens once, at import. Reading is free. Asking is navigation.
Drop any file or connect the document management system. Hippocampus will ingest your documents, keep the access rights intact and understand each document. Documents will become a wiki of the company knowledge factory.
Browse wiki with your company data, make changes, read summaries and navigate table of contents. You can always see the source of the information. Wiki is made easy for humans and handy for AI to be quick and efficient.
Choose a scope to a folder, document or the whole wiki - ask using LLM in plain language and get a trustworthy reply based on your documents. LLM-powered wiki is the most powerful tool for your daily productivity you can imagine.
The problem
Traceability, efficiency, and knowledge that never leaves one chat window. Each has a specific answer.
Assistants paraphrase. When the auditor asks “which clause, which page”, a plausible summary is worth nothing.
Hippocampus: every answer quotes the passage with its page; a question the documents cannot answer is refused.
Full context is precise but slow and expensive. Chunked vectors are cheap until the answer needs structure. Reasoning at every question does not scale.
Hippocampus: understanding is done once at ingest, so a question is navigation, not re-reading.
Answers in a personal thread do not compound. Teams need the same truth in the tools they already use.
Hippocampus: workspaces with permissions, MCP and API on every tier, open Markdown you can export.
Against the alternatives
Quality, speed, cost, scalability, traceability. Hippocampus is the dotted outline; pick an alternative to fill the chart.
Their strengthFree and friendly for one person's notebook of sources.
Strongest disadvantagePer user, with notebook and source caps; no team permissions or tenants; citations at source level and no measured refusal.
NotebookLM: Generous free tier bundled into Google's consumer plans; a personal product by design.
Hippocampus: Quality: cited, checkable answers across documents. Speed: bounded retrieval with a latency ceiling; a page imports in seconds. Cost: understanding is paid once at ingest; a published per-credit tariff and a free tier. Scalability: workspaces with permissions, shared or dedicated EU stacks, field seats, MCP and API on every tier. Traceability: byte-exact quotes with the page and the original; refusal when the documents do not say.
Their strengthTemporal memory for conversational agents — strong on what a user said, and when.
Strongest disadvantageBuilt for chat memory, not documents: credit-based pricing, eventually-consistent ingest, facts rather than cited passages.
Zep: Ingestion-metered credits from a three-figure monthly plan; self-host edition discontinued 2025; benchmarks are conversational, not document Q&A.
Hippocampus: Quality: cited, checkable answers across documents. Speed: bounded retrieval with a latency ceiling; a page imports in seconds. Cost: understanding is paid once at ingest; a published per-credit tariff and a free tier. Scalability: workspaces with permissions, shared or dedicated EU stacks, field seats, MCP and API on every tier. Traceability: byte-exact quotes with the page and the original; refusal when the documents do not say.
Their strengthCheap, quick indexing of a single long PDF with page-level provenance.
Strongest disadvantageNo passage quotes or refusal record; cross-document routing is a known weak spot; query-time reasoning is slow; no team or tenant model.
PageIndex: Per-page indexing with credits; vendor-run benchmark figures; a developer API rather than a product a team signs into.
Hippocampus: Quality: cited, checkable answers across documents. Speed: bounded retrieval with a latency ceiling; a page imports in seconds. Cost: understanding is paid once at ingest; a published per-credit tariff and a free tier. Scalability: workspaces with permissions, shared or dedicated EU stacks, field seats, MCP and API on every tier. Traceability: byte-exact quotes with the page and the original; refusal when the documents do not say.
Their strengthEnterprise search across every system with permissions inherited and SSO everywhere.
Strongest disadvantageSearch results and links rather than a checked, quoted answer; seat pricing with large minimums puts it out of reach below a large organisation.
Glean: Per-seat pricing with contract minimums; connectors and permission inheritance are the product.
Hippocampus: Quality: cited, checkable answers across documents. Speed: bounded retrieval with a latency ceiling; a page imports in seconds. Cost: understanding is paid once at ingest; a published per-credit tariff and a free tier. Scalability: workspaces with permissions, shared or dedicated EU stacks, field seats, MCP and API on every tier. Traceability: byte-exact quotes with the page and the original; refusal when the documents do not say.
Their strengthOpen-source extract–cognify–load pipeline into a knowledge graph, good for domain schemas.
Strongest disadvantageWrite-side LLM cost and slow graph traversal; graph triples rather than cited passages; a developer framework to operate, not a product a team signs into.
Cognee: Open source with a metered cloud; practitioner notes on write-side cost and traversal latency.
Hippocampus: Quality: cited, checkable answers across documents. Speed: bounded retrieval with a latency ceiling; a page imports in seconds. Cost: understanding is paid once at ingest; a published per-credit tariff and a free tier. Scalability: workspaces with permissions, shared or dedicated EU stacks, field seats, MCP and API on every tier. Traceability: byte-exact quotes with the page and the original; refusal when the documents do not say.
Why teams choose Hippocampus
What
One question over many documents, answered from the one that holds the answer — and shown.
How
Each document is understood once into a summary graph whose nodes point to byte ranges in the source; the answer carries the passage, the page and the file.
What it means for you
No more “which file was that in” — and nothing stitched from the wrong contract.
How answers are built →What
When the documents do not say, the answer says so.
How
Nothing not drawn from a retrieved source reaches the answer; “nothing found” is absence, not licence to guess. The behaviour is checked on every release.
What it means for you
A refusal is information: widen the scope, or import the missing document.
Against the alternatives →What
Cost sits at import; asking stays light at any volume.
How
Two retrieval funnels — top-down over the structure, bottom-up over the finest sections — run in parallel and fuse; retrievals per question are capped.
What it means for you
Ask as often as the work needs. The bill follows what your team asks, not how much a model read.
The tariff →What
Designed to survive compromise: the architecture bounds what an attacker gets.
How
Every document has its own key, bound to the workspace and wrapped in a KMS; access control runs inside retrieval, before ranking; every vault operation is logged in sequence.
What it means for you
The security review is passed before the price conversation starts.
Security and data handling →What
The price of the next action is shown before it is taken.
How
Pages, questions, seats and workspaces are priced in credits on one published tariff; the meter is visible and fails closed.
What it means for you
No surprise on the invoice, and no per-seat lock-in below Corporate.
Price list and estimator →What
The knowledge base is not locked in a proprietary index.
How
Every document becomes an Open Knowledge Format bundle; the MCP server and the REST API give agents and scripts the same corpus under the same access control.
What it means for you
Switching cost stays low on purpose. The moat is the answers, not the lock.
Deployment options →Deciding
The technical reasons to choose it, and the goals the business reaches with it.
Technical
Business
No card. No API key. Your first month starts with a welcome grant of credits.