AI Enterprise Operating Environment · Self-hosted · Enterprise
Run AI Workbench across the whole organisation without losing sight of it.
The AI Operating Environment is the control centre your AI work reports to. It shows every workbench and agent node in one place, applies your policies before anyone gets the chance to work around them, shares approved skills, roles and prompts with every team, and tells you what AI is costing you and who is spending it. It runs in your own cloud or data centre, so your data stays where your policies say it must.
- Entra ID first, any OpenID Connect issuer
- Kubernetes, Azure or on-premises
- Same images as Mojo Up AI Cloud
Governance, observability, cost
Replace opaque chat with visible activity, plans, tasks and roadmaps.
A portfolio manager files projects under business areas. Work on them is picked up by AI Workbench teams, people, nodes and CI runners. Every run is traceable to a person, a licence, a machine and a project.
One view of the fleet
Every workbench, node and agent in your organisation, who is working on what, drill-down to the transcript and the diff, and a pause button on one screen.
Policy every workbench applies on its own
Which agent tools, providers, approvals, tool servers and network destinations a repository allows. A ceiling no settings layer can loosen, rendered from each workbench build's own manifest.
An organisation catalogue
Approved skills, roles, prompts, tools, teams and design systems, published once and pulled by every workbench. MCP tool servers reviewed and set to Allowed, Ask first or Off.
Spend, budgets and chargeback
Token usage and compute priced and attributed to person, team, project and business area. Budgets that warn, throttle and block, with an audited human override. Monthly chargeback as CSV or API.
The enterprise brain
Shared memory from every workbench and project merged into one knowledge graph, with access following the portfolio and compartment rules. Every result names its source.
Approvals on the phone
Enrolments, node joins, filing requests and budget overrides approved from the console or the mobile app, with the same audit record either way.
Nodes and compute pools
Headless workbenches on shared compute, approved by a person with the right role, assigned to teams, with a project's ready tasks scheduled across desktops, nodes and CI runners.
Secrets by name
API keys and tokens kept at organisation, team, project or repository scope, sealed at rest under a key in your vault, released only to a signed-in machine for a member, with an audit row each time.
A hash-chained audit trail
Every approval, enrolment, dispatch, task hand-off and policy change in an append-only, hash-chained log naming the person, agent and compute involved. Search, export and verify it from the console, or forward it.
Built for Microsoft 365
Your directory, your labels, your SIEM.
- Entra ID
- People sign in with the organisation's own tenant. App roles carry authorisation. A plain member sees and controls only the workbenches they enrolled.
- Agent 365
- An agent identity and a registry card for each enterprise team, and every finished run exported to Agent 365 observability.
- Purview
- Sensitivity labels and their markings flow into policy; Purview's DLP verdict applies to what an agent is about to send; label decisions are reported as Purview activity.
- Sentinel
- The audit trail forwards to your SIEM, so AI work sits beside everything else your security team watches.
- Work IQ
- Agent memory grounded in Microsoft 365, through the organisation's own Work IQ.
Deploy anywhere
The same containers, three ways to run them.
Stateless API behind Postgres 16 and Redis, a static console, optional object storage for encrypted backups and optional mesh services. No cloud SDK inside the services.
Azure
Bicep templates for Container Apps, PostgreSQL Flexible Server, Azure Managed Redis, a registry and Log Analytics. The same layout Mojo Up runs.
Kubernetes
A Helm chart for any cluster: bring your own Postgres and Redis, run the API with as many replicas as you like.
On-premises
docker compose as the reference layout. Air-gapped installs with the licence check-in turned off.
| Basic | High availability | |
|---|---|---|
| For | A pilot, a small organisation, staging | Production people depend on |
| Replicas | One API, one console | Two or more of each, across zones |
| Postgres | One server | Zone-redundant standby, 35 days point-in-time restore |
| Network | Public endpoints | Private endpoints, optional WAF |
| Order of cost on Azure | US$50 to 260 a month | US$500 to 1,000 a month |
Questions
What security and platform teams ask.
Bring your AI work under one roof.
Tell us about your organisation and we will walk you through a deployment. Or pull the images and start the 30-day trial today.