About Memtro
Memtro is company memory for the AI tools a team already uses: one endpoint that knows the organisation, reads its systems, and answers with its own facts cited.
Why it exists
Most company AI fails for a dull reason: the model does not know the company. It cannot see the CRM, the ledger or last week's thread, so every answer starts from nothing and every person pastes the same context into every prompt. Memtro sits in front of the models a team already pays for and gives them the company: a shared memory of people, customers, decisions and promises, and read access to 144 systems through tools that cite the record they read.
What we believe
- An answer without a citation is a guess. Every figure Memtro gives points at the record it came from.
- Read first. Connectors read by default; a tool that changes something is marked, used only when asked, and granted by name.
- Your keys, your models. Memtro routes requests across the providers you hold keys for and never resells tokens.
- Memory has a timeline. What the company believed in July is a different question from what it believes today, and both should be answerable.
- One price per organisation. Connectors, jobs, memory and keys are never metered, so nobody is punished for using it.
How it is built
Memtro is an OpenAI-compatible endpoint, an Anthropic Messages endpoint and an MCP server over one memory graph, hosted in the United Kingdom. It is built by a small team and shaped by the organisations using it: the changelog is the honest record of what ships and why. How it treats data is set out under Security, Privacy and Subprocessors.
Get in touch
For a conversation about your organisation, write to sales@memtro.com or see the contact page. To try it, start a free trial: a model key and a first connector take ten minutes.