Ask naturally, get real answers.
Minerva interprets what you actually mean — rephrasing and routing each question before it answers — so you don't have to speak "search engine."
Wisdom, on demand
Private AI for the knowledge you can't send outside.
Minerva is a self-hosted AI assistant that answers from your own documents, data, and code — every answer grounded, cited, and kept entirely inside your own walls.
Pilot in 4–6 weeks with a selected set of your own document sources.
Built for EU regulation — AI Act, GDPR & beyond →minerva.cmas
Ask Minerva…
What's our data-retention policy for customer records?
Customer records are retained for 24 months, after which they're automatically archived to cold storage and purged after a further 12 months. [1][2]
Verified · grounded in 2 sources
1 · Data Policy2 · Compliance Handbook
Built for regulated organisations — telecoms, utilities, insurers, financial services, and public sector.
Your answers already exist — buried in documents, databases, and code across multiple systems. Finding them shouldn't take an expert and hours of digging.
Why Minerva
Most AI assistants ask you to send your knowledge somewhere else. Minerva is built the other way around.
Typical hosted AI
- Your data leaves your network
- Answers you can't trace back
- Flat keyword or vector search
- The vendor controls your access
- Locked to a single provider
Minerva
- Everything stays inside your walls
- Every answer cited to its source
- Graph-aware — it connects the dots
- You control the infrastructure and access
- Portable knowledge, no lock-in
What Minerva does
Imagine the colleague who has read everything, forgets nothing, and answers in seconds — always in context, and always with receipts.
It connects the dots.
A knowledge-graph layer maps the relationships between people, systems, and concepts in your data — answering questions a document search never could.
Answers you can trust.
Every response is grounded in your own sources and cited, with built-in guardrails that keep Minerva honest instead of guessing — and that get sharper with every iteration of the product.
One brain, many domains.
A single platform serves multiple knowledge areas and connects to the systems you already use — from databases to documentation to codebases.
How it works
From question to grounded answer
01
Ingest & connect
Minerva indexes your sources and builds both a semantic index and a knowledge graph of how everything relates.
02
Understand the question
It analyses and refines each query, then chooses the best retrieval strategy — hybrid keyword + meaning search, or graph traversal.
03
Answer & verify
It generates a grounded answer, checks it against the retrieved evidence, and shows its sources.
Minerva's retrieval and ingestion pipelines combine classical and neural techniques, with verification at the end so answers stay grounded:
Ingestion components adapt to whatever sources you have — or are tailored to a specific source when it has particularities worth handling directly.
Hybrid BM25 + vector retrievalModel-aware embedding strategiesQuery rewritingClassifier-based routingCross-encoder rerankingGraphRAG-style knowledge graphGrounding verification
Use cases
One platform, wherever knowledge lives
Internal knowledge assistant
Instant answers from your docs, wikis, and policies.
Technical & code intelligence
Understand large codebases and systems through natural questions.
Data-grounded Q&A
Query operational data and documentation in one conversation.
Deployment
Deploy where your data already lives
Minerva runs entirely inside infrastructure you control — connected to the documents, databases, and code your teams already use. No data leaves your boundary.
On-premises
Run Minerva in your own data centre, fully under your operational control.
Private cloud / VPC
Deploy into your own cloud tenancy — your region, your network boundary, your keys.
Air-gapped
Operate fully offline with no outbound connectivity, for the most sensitive environments.
Yours to keep
Control your knowledge internally — don't rent it from a platform that can cut you off or move your data out of reach.
Some knowledge is too sensitive and too tightly governed to hand to an outside model. Minerva keeps your documentation and the intelligence built on top of it within your own infrastructure — your sources, your index, your environment.
You always know exactly where your information lives, who can reach it, and that no third party can read it, train on it, or decide when you get access to your own answers.
Built for EU regulation
Ready for the AI Act — and what comes next
Because Minerva runs entirely on infrastructure you control, the hardest parts of EU compliance — knowing where data lives, who can reach it, and how every answer was produced — are built into the architecture from the start, rather than bolted on afterwards. The same foundation maps cleanly onto regulations already in force and those still taking shape.
EU AI ActIn force · phasing in
Transparency, oversight & records
Cited, traceable answers and full retrieval logs give you the transparency and record-keeping the Act expects — while human-in-the-loop review and self-hosting keep meaningful oversight and data governance in your hands.
GDPRIn force
Data stays where it belongs
When deployed in the standard self-hosted configuration, personal data never leaves your infrastructure and is never shared with or used to train outside models — so data residency, purpose limitation, and the right to know where information lives are addressed by design.
Data Act · DGAIn force
Sovereignty over your data
You retain full control and portability of the knowledge Minerva indexes. No platform lock-in, no third party deciding access — aligning with the EU's data-sharing and governance principles.
NIS2 · DORAIn force / phasing in
Security & operational resilience
Running on infrastructure you control means your existing security posture, access controls, and audit trails extend straight to Minerva — supporting cyber-resilience and operational-continuity duties.
Upcoming EU actsOn-going
A foundation that adapts
As new EU rules are published, the same architecture — in-house data, traceable answers, configurable controls — is the groundwork to adapt quickly rather than re-platform.
Common questions
No. When deployed in the standard self-hosted configuration, Minerva runs inside your own infrastructure, so your documents and the answers built on them never leave your environment — and are never shared with, or used to train, an outside model. The data-transfer and processing concerns that usually trigger GDPR and AI Act scrutiny are far less likely to arise.
GDPR · EU AI Act
That depends on your use case, not the tool alone — but Minerva is built to keep you on the right side of it. Human-in-the-loop answers, full traceability, and logged sources give you the transparency, oversight, and record-keeping the Act asks for, so you can demonstrate responsible use rather than scramble to retrofit it.
EU AI Act
Yes. Every answer is grounded in your own sources and cited, and the retrieval steps behind it are logged. You get an auditable trail from question to evidence to response — exactly the kind of explainability and record-keeping the AI Act and sector regulators expect.
EU AI Act · DORA
When deployed in the standard self-hosted configuration, your data stays exactly where you put it — your region, your data centre, your cloud tenancy. You always know where information lives and who can reach it, which is the heart of GDPR data residency and the EU's data-sovereignty rules.
GDPR · Data Act · DGA
Unlikely. The same foundations that cover today's regulations — in-house data, traceable answers, configurable controls — are the groundwork to adapt as new acts are published, so you adjust configuration rather than rip-and-replace.
On-going EU acts
Assess your exposure
Check your AI Act readiness
See how the EU AI Act affects your organisation's activities — today and as you plan ahead.
Minerva is an enabling platform: it gives you the controls, traceability, and data residency that compliance demands. This questionnaire is a plain-language guide only — it supports your obligations and does not replace legal review or certify conformity on its own.
Shipping continuously. Minerva is a CMAS platform with an active roadmap — sharper retrieval, new connectors, and steady production improvements landing on a regular cadence.
Now accepting design partners
Help shape private AI for regulated knowledge.
We're working with a small number of selected organisations to validate Minerva in real enterprise knowledge environments — policies, technical documentation, support knowledge, codebases, and operational data. Design partners help shape the roadmap while testing private, cited AI inside infrastructure they control.
Apply to the design partner programmeBuilt by CMAS for environments where trust isn't optional — controlled deployment, source-grounded answers, traceability, and portability from the start.
Security & trust
Runs in your infrastructureYour encryption, at rest and in transitConfigurable data residencyYour access controls & audit trails
Certification status (e.g. ISO 27001, SOC 2) available on request — we'll share where we are today.
Get insights from
what you already own.
We work alongside your organisation to make your information and knowledge more powerful — turning what you already have into answers you can act on.
Pilot in 4–6 weeks with a selected set of your own document sources.
Thanks — your request is in.
We'll be in touch within one business day to set up your walkthrough.