FOR TEAMS PUTTING AI TO WORK ON REAL SYSTEMS
AI proposes. People decide.
Connect an AI to OpenX and it can build: connections, queries, access, removals. Over MCP, every change it makes is a proposal a person accepts. Most can be undone until Deploy, and only a person deploys.
New to OpenX? Key terms →
FROM AN AI'S IDEA TO WHAT RUNS
- 01 · Propose
- The AI drafts a change over MCP
- 02 · Check
- OpenX tests it and says why if not
- 03 · Accept
- A person accepts it in the product
- 04 · Undo
- Most accepts can be taken back until Deploy
- 05 · Deploy
- A person ships it into your own Azure
Save freely. Deploy deliberately.
WHERE THE PERSON STAYS IN THE LOOP
Review every change before it reaches your cloud.
A save, like a new query, serves as soon as it is saved. Deploy is the step that builds or removes resources in your Azure, and every removal waits for it.
Proposals, not edits
An AI connected over MCP changes nothing directly. It proposes, OpenX checks the proposal, and a person accepts it.
Access an AI over MCP can suggest, never grant
An agent connected over MCP can propose an access right, an access card or a job identity. It waits as Proposed until a person accepts it, and the agent never handles secrets.
One Deploy window
Everything Deploy will build or remove in your Azure, listed before it happens, with how each removal will be handled.
Removals you can take back
A removed node greys out with Restore until Deploy. Protect from deletion keeps a data store from being removed through OpenX at all.
Live and edited, kept apart
History separates Deployments from Saves, and a workspace never runs two deploys at once.
A record outside the AI's reach
With Provenance on, the record says who did what, resolved by the server, and no agent can edit it.
IF "THE AI DID IT" IS NOT AN ANSWER
For teams that want AI's speed without losing the say.
Teams adopting AI agents
Let agents draft real changes to real systems, with a person between every draft and what runs.
IT owners
An automation or an AI cannot quietly delete cloud resources: removals are held until someone deploys.
Regulated teams
Show what changed in your cloud, when, and who pressed Deploy, with a record the AI cannot touch.
NEXT TO OTHER HUMAN-IN-THE-LOOP TOOLS
Human approval is table stakes. The reach of the review is not.
Deploy tools preview a stack, and agent platforms ask before risky actions. OpenX puts every AI change behind a person, and every change to your cloud behind one Deploy, across the whole workspace. One exception, by choice: switch on Greenfield mode and Azi, OpenX's built-in assistant, can accept its own drafts. Removals and job identities still wait for a person, and Azi can accept only what the person who switched it on could, access changes included.
- One Deploy for every cloud change in the workspace, not one stack's or one site's.
- A removal is held, not done, until a person deploys.
- It deploys into your own Azure, with the record kept there too.
WHAT THE REVIEW LOOKS LIKE
An AI does the work. A person signs it off.
A clean-up you can trust
- 01An agent proposes removing connections nobody uses.
- 02They grey out with Restore.
- 03A person reviews and deploys, and Provenance records it.
Onboarding a new team
- 01An agent connected over MCP proposes access cards for the team.
- 02A lead accepts them in App Identity.
- 03The API and the MCP server honour them.
A protected system of record
- 01A Cosmos store is set to Protect from deletion.
- 02An agent proposes removing it.
- 03The removal is refused until a person turns protection off.
WHERE CHANGE CONTROL IS THE JOB
Change on your schedule, not the AI's.
LIFE SCIENCES
Change control on a monitoring system, with a reviewable record of each deploy.
MANUFACTURING
Stage cloud changes during the week and deploy at a shift change.
ENERGY AND BUILDINGS
Let an assistant draft changes across sites, and ship them when a person is ready.
Let AI build. Keep the last word.
Connect any MCP-capable AI to your workspace and review everything it proposes.
MORE OF OPENX
OpenX MCP Server →
A few tools let any MCP-capable AI read your workspace and propose changes to it.
API Manager →
Your workspace is already an API. Decide who may call it and what each caller reaches.
Warm queries →
Save a question over live data and serve the answer to dashboards, apps and AI.
Provenance →
A record you can prove, kept in your own cloud, with corrections and signatures.
Your cloud →
Connect your Azure, build the landing zone, and the rest of OpenX runs inside it.
Databases →
A Cosmos DB store in your subscription, with a governed API for each container.
App Identity →
Your own sign-in directory, identities for your jobs, and access in one window.
Licenses →
Sell what you build on your own Stripe, and turn a subscription into a plan's rights.
IoT →
IoT Hub and Data Explorer in your Azure, feeding queries, APIs and AI.
Communications →
Send mail through your own relay, with the password in your vault and sends on record.
App building →
Data, sign-in, access, an API and billing: your app's backend, already there.