AI first?We put youSecondFirst.
AI-native systems built around work already in motion.
We design, advise on and deploy systems where people and AI work from the same reality.
01 / What we do
What we build
Three ways in. All of them end in software that runs — not a deck.
- 01
AI Solution Development
We design and build AI-native systems around the workflow itself: the records, exceptions, roles and approvals that give the work meaning. People and agents work from the same operating reality. Recent builds: a hosted platform where AI agents create persistent apps, and the WebXR education-analytics platform used in a live university research collaboration.
- 02
AI Consulting & Advisory
Decide where AI belongs before adding it. We map the real constraint, identify what should remain human and define the boundaries before implementation begins. You get a written map you can disagree with — not a vision statement.
- 03
Deployment & Operations
Self-hosted or cloud. Staging before production, rollback paths, audit trails, and a clean exit with your data. We run our own platforms on the same discipline we sell.
02 / How it works
One reality.Not another handoff.
An agent should not be asked to guess what an organisation already knows.
We connect the same operating context across the interfaces used by people and AI. The work does not have to be rebuilt in a separate system before it can move.
- State
- Context
- Rules
- Permissions
- Actions
In practice: an agent's action goes through the same permissions and approvals as a person's. A lock held by one user refuses another — quietly, as state, not as an error. Every consequential step waits for explicit confirmation, and everything that happened stays on the record.
03 / What stands behind it
On the record
What stands behind it is not a claim — it's on the record: an open-source engine published to npm and the official MCP Registry; a hosted agent-app platform in public beta; an iOS app on the App Store; and the WebXR education-analytics platform behind a live university research collaboration. All of it inspectable. All of it running.
- Open MCP Apps — MIT, on npm
- openmcp.app — public beta
- Gretel — App Store
- University research platform — in production
04 / Where to begin
Start wherethe friction is.
Bring the part of the work that is repeated, delayed, opaque or ready to change. You do not need an AI brief.
We begin by finding the smallest useful move: what needs to stay human, what can become assisted and what must wait for a clear decision.
- Keep human
- Judgement, approval, responsibility and the right to change the path.
- Assist
- Search, preparation, routine updates and bounded execution with context intact.
- Wait
- Uncertain, irreversible or consequential actions pause for explicit confirmation.
A useful next move
Bring the work, not the hype.
Tell us about a live workflow, a difficult exception or a decision that keeps being rebuilt from scratch. We begin with the reality already there.
Start a conversationSecondFirst