Infrastructure for AI-native services
Theia Craft builds the AI Operations Platform: the foundation for delivering professional services assisted by agents.
Let's talkWhat we work on
AI Operations Platform
Three layers between the conversation and the systems.
Models and tools stay behind controlled interfaces: they can be replaced without rebuilding the system.
The difference between an assistant and an infrastructure lies in what you can answer afterwards.
About any past operation, in front of a client or an auditor.
Who authorised this operation?
The identity that ran it, its autonomy level and the approval are part of the record.
What data and sources did it use?
The context and the documents involved are kept together with the result.
Which version of the procedure ran?
Every procedure is versioned: you know which one acted and within which limits.
What did it produce and who validated it?
The output, its validation and who signed it off.
Governance
Every operation has a level of autonomy.
There are no generic tools. An operation is declared with its purpose, its output format, its boundaries and how far it may go on its own.
Direct
Reading or calculation over already validated information.
Draft
The system writes it up and waits for professional review.
Staging
What is contributed stays aside until someone accepts it.
Approval
Nothing runs or closes without explicit authorisation.
By default everything requires human validation. Operating without it is earned through tests, a version and an authorisation that can be revoked.
Lifecycle
From a real case to a tested procedure.
The platform does not learn by acting on its own. A procedure is observed in real work, formalised and only then enabled.
01
Real case
02
Pattern
03
Design
04
Tests
05
Approval
06
Rollout
07
Monitoring
The cycle does not end at rollout: monitoring decides whether a procedure stays, gets adjusted or is retired.
How we build
Six principles.
- P-01
AI amplifies the specialist
It extends the professional; it does not own the judgement.
- P-02
Minimum friction
The client never administers AI infrastructure.
- P-03
Data before prose
Structured information, not closed answers.
- P-04
Structural isolation
Separation between clients never rests on a prompt.
- P-05
Proportional cost
Each component arrives when there is real usage.
- P-06
Service before product
Operation decides what deserves to be a platform.
In production
Prufy already runs on the platform.
Managed technology risk and audit service.
The studio
Engineering with an auditor's judgement.
Theia Craft has worked since 2025 as an applied AI engineering and R&D studio. It builds the platform and learns from how services operate on it. Prufy came later, within the studio.
Kiyoshi Omaza
Founder
Technology risk and cybersecurity. IT internal audit, cloud security and regulated environments at Grupo Cajamar, Deloitte, EY, Accenture, NTT DATA and Fujitsu.
LinkedIn