- Keep your provider's API: a new base URL and two headers
- A tenant, a plan and a policy for each of your customers
- A CLI, TypeScript and Python SDKs, an MCP server and an HTTP API
- Your own model keys and private endpoints
AI infrastructure, private by design
The complete pipeline for private AI.
Privacy, memory, context, masking and routing, built to work as one. Build your own AI products on it, from the command line, the SDKs or your agents.
In development. Nothing is available yet.
The pipeline
Five stages. One pipeline.
Each stage does one job and sees only what that job needs. Together they keep personal data in scope from the question to the answer.
01Privacy
Who may, and where to
Identity and a signed policy that can only narrow.
Brain Authority02Memory
Kept with its source
Documents, notes and conversations, in Spaces kept apart.
GenesysBrain03Context
Only what this request needs
Recalled from what the caller may read, each piece with its source.
GenesysBrain04Masking
Details become tags
Names and numbers are tagged before anything leaves; the values stay sealed.
Masking service05Routing
The right model, in bounds
A model and region the policy allows. The decision is kept, not the words.
Brain Ways
Why one pipeline
Privacy breaks in the gaps between tools. So we built the whole path.
Masking means little if the next service logs the prompt. A router cannot know what a memory may share.
Here each stage is built for the next: the router takes masked-only traffic only with a signed claim from masking, and keeps no content.
The infrastructure in fullFollow one request
One message through all five stages.
What each stage does, and what it sees.
For developers
Build on the pipeline, from any tool.
The same operations from the command line, the SDKs, an MCP server and the HTTP API.
02CLI
Drive the whole pipeline from the command line.
Every operation, for people at a terminal and for their scripts. In development.
The CLIgb auth login
Signed in. The session stays in your keychain.
gb memories ingest --document contract.pdf
{"jobId":"job_…","status":"Accepted"}
gb mcp serve
The same operations, as tools for your agents.
By design
Four things we build in, not bolt on.
01Masking
Masked before the model
Names and numbers become tags. The model sees tags; you read the answer in clear.
02Retention
No content on the route
The router keeps no request or response. Only the decision is recorded.
03Training
Never trained on
Your content never trains a model, ours or anyone else's. We do not sell it.
04Provenance
Traceable to its origin
Every record keeps its source. Every route keeps its reason.
Built for the rules
Designed around the EU AI Act and the GDPR.
Built from the start for the strictest privacy rules.
| Law | What it asks of an AI product | How we design for it |
|---|---|---|
| EU AI ActRegulation (EU) 2024/1689 | Tell people they are talking to an AI; mark generated text in a machine-readable way; support AI literacy. | Provenance on every record and a record of every routing decision. Each product built on it discloses itself and marks its output; the pipeline keeps the records it needs. |
| GDPRRegulation (EU) 2016/679 | A lawful basis, minimisation, security, the rights of access and erasure, safe transfers. | Masking before any model, Spaces that keep data apart, provenance on every record, erasure by Space. |
| ePrivacyDirective 2002/58/EC | Consent before storing or reading information on your device, unless strictly necessary. | This site sets no cookies; it stores your theme and language, only when you choose them. |
| US state privacy lawsCalifornia and others | Notice, access and deletion rights, opt-outs of sale and sharing. | We do not sell personal data or share it for advertising. |
Progress · November 2026
Where we are.
Where each stage of the pipeline stands.
| Part | State | Where it stands |
|---|---|---|
| PrivacyBrain Authority | Built | Version 1 complete and tested. |
| MemoryGenesysBrain | In progress | Ingest and sessions are built and tested. Erasure is in development. |
| ContextGenesysBrain | In progress | Retrieval in scope is built and tested. Search and conversations are in development. |
| MaskingMasking service | In progress | Detects personal details in English, with more languages to follow. |
| RoutingBrain Ways | Built | Version 1 complete, tested for resilience, load and security. |
| Developer toolsCLI, SDKs, MCP | In progress | The CLI, the SDKs and the MCP server are built and tested for ingest and keys. |
| HostingWhere it will run | Planned | Infrastructure and providers are being selected, with EU model endpoints planned. |
BuiltIn progressPlanned
Everything has an origin
Private AI starts with the pipeline under it.
Before launch we publish the documentation, the providers and the terms that bind us.