01
Every session starts from zero
In the morning you explain the client, the project and the background to your AI. In the afternoon, in the same tool, you explain it all over again.
Operated in Nuremberg
Your AI forgets everything.Not anymore.
knowmind is the body of knowledge that Claude, ChatGPT, Cursor and your own applications read from and write to together. You file a note once, and every tool knows it in every session that follows.
Up to 2,500 memories free for good · no payment details · exportable as a file at any time
- Operated in Germany
- Servers at Hetzner Online GmbH in Nuremberg, in a data centre certified to ISO 27001. The operator is a German company with no US parent.
- The indexing stays here too
- The model that turns your texts into searchable vectors runs on the same server. No external service is called when saving.
- You can leave at any time
- Your body of knowledge, tenant data and audit log are ready for export. Cancelling takes one button, no sign-in and no conversation.
knowmind works with
ClaudeChatGPTCursorClaude DesktopClaude Coden8nAnthropic ClaudeOpenAIGoogle GeminiMistral AIAleph AlphaAzure OpenAIOllamaLM StudiovLLMModel Context Protocol®
Named here are tools and models that can be connected. All trademarks belong to their respective owners; naming them points to compatibility and implies no business relationship, endorsement or certification. All ways to connect
The problem
Your company works with AI every day and learns nothing from it.
The knowledge from your work stays nowhere. It sits in chat histories nobody finds again.
02
Every tool is its own silo
What you discussed in Claude, Cursor does not know. What sits in a ChatGPT history, nobody finds again. Knowledge falls apart along tool boundaries.
03
The store belongs to the provider
The memory features of the large providers sit in their systems, in their format, under their law. Switching means starting over.
How it works
What happens when your AI asks.
Six steps, from question to answer. It makes no difference which tool asks: the body of knowledge behind it is the same. knowmind does two things here without being asked. It stays silent on questions with no substantive anchor instead of padding the context with chance hits, and it turns away credentials at the door when an agent tries to write a password into the memory.
Your tool asks
Claude, ChatGPT or Cursor asks a question. All three use the same access and the same body of knowledge.
The question is read for what it is about
A model on our own server works out what the question concerns. No external service is called for this; your text does not leave the machine.
Three searches run at once
One for the words themselves, one for the meaning, and one across the links between people, projects and decisions. Together they find passages a single search would miss.
The result lists become one
Whatever ranks high in several searches rises. That way a chance hit from a single method does not win.
What genuinely fits remains
A second model checks every hit against the question. Out of thirty candidates, the few your AI actually needs come through.
Answer returns, insight stays
Your tool receives the hits with their source. Whatever new emerges is filed in the same body of knowledge.
Measured
knowmind finds four out of five questions right away.
Measured on twenty questions from everyday work against 1,219 memories, on 23 May 2026. The questions were held back while tuning the search so the result stays sound. The comparison is against the methods running in your company today.
| Method | Hits within the top five | Context per answer | Response time |
|---|---|---|---|
| Full-text search across files | 55 % | 3.244 | 27 ms |
| Search by file name | 55 % | 2.660 | 2 ms |
| Vector search alone | 65 % | 1.620 | 491 ms |
| knowmind | 80 % | 1.790 | 475 ms |
Scripts, question set and raw data are published, so the measurement can be recalculated. See method and raw data
Evidence
From 2 August 2026 you will be asked what your AI based itself on.
That is the day the transparency obligations under Article 50 of the AI Act take effect. For high-risk systems, Article 12 requires automatic logging across the entire lifetime. Anyone who has to show which state of knowledge a decision rested on needs a store that knows its own past.
Every entry carries two dates
Since when something holds true, and since when knowmind has known it. If something changes, the old entry is marked superseded and stays retrievable. Nothing is deleted merely for being out of date.
Every change appears in your workspace audit log, 30 days to 5 years depending on plan, exportable as a file.
Payment terms for client M.?
As of 14 March 2026
→ 30 days
As of today
→ 14 days
changed on 2 June 2026, previous version kept and dated
How far the timeline carries, and who in your workspace may change entries, is set out with sources in the Trust centre.
Access
A memory is only as sovereign as the law its operator answers to.
Server location alone says little about that. What counts is who can be served an order when it comes to it.
The CLOUD Act does not ask about location
The US statute obliges providers subject to American law to hand over data within their control. The text of the law states explicitly that the place of storage makes no difference.
Asked, and not guaranteed
Before the French Senate in June 2025, the legal counsel of Microsoft France was asked whether he could rule out French data being passed to US authorities. He could not.
In our chain there is no addressee
knowmind is operated by a German company with no US parent. The data sits in Nuremberg, and indexing happens on the same server.
What this means for your requests to Claude or ChatGPT
Your request still goes to the provider you chose, and therefore into that provider's legal space. We cannot change this, and we do not claim otherwise. What is handed over are the five to ten memories that match the question. The body of knowledge itself, all relationships and the entire history stay in Nuremberg. If you connect a model in Europe or one you run locally, the request does not leave European jurisdiction either.
The EU Data Act has applied since 12 September 2025. It obliges providers of data processing services to offer open interfaces and to hand over data in a machine-readable format; from 12 January 2027 switching charges are banned outright. knowmind connects through the open Model Context Protocol. The same connection carries Claude, ChatGPT, Cursor and local models.
Every claim on this page can be verified
The trust centre lists every claim on this page with its source: hosting and data centre, tenant separation, backups and the last restore test, subprocessors and the data processing agreement in full. The technical version for your IT department sits alongside it, with the details a security review needs.
Getting started
Set up in ten minutes, no developer skills needed.
Create an account and write
File notes and documents straight into knowmind and search them in the dashboard. No setup, no further software.
First stepsConnect once, read everywhere
Claude, ChatGPT, Cursor, Gemini and Cline reach the same body of knowledge over one standardised connection. In claude.ai and ChatGPT, storing the access once is enough.
Connection guideConnect your own applications
A documented interface and a command-line tool that reads in whole directories. Included in every plan.
Interface referenceHow does existing knowledge get in?
Through the interface you upload documents one by one. With the command line tool you read in whole directories, where identical content is recognised by its checksum and not stored twice. Ready-made connections to SharePoint, Outlook or a document management system do not exist today; migrations like that run through the interface and are project work. If that is your case, talk to us before you start.
Plans
Private stays free, Pro starts at 15 euros a month.
On the Private plan you file up to 2,500 memories, with no payment details. Pro lifts the limit. Team brings a shared body of knowledge for five people with roles and a data processing agreement; Business is the larger tier with ten seats and longer retention of the audit log. Cancellable monthly.
The knowmind client (CLI and MCP server) is open source under Apache-2.0; the hosted platform is a proprietary service.