Dextre on Azure
In development · capabilities below are planned, not yet live
Dextre takes a request for an Azure deployment, prepares the infrastructure code, shows a named approver exactly what will change, and deploys only what that person approved. Every step leaves a record the client can keep.
How a deployment moves through Dextre
The order is fixed and a person sits at the centre of it. These six steps are the approval flow for Dextre on Azure.
Request
A structured ask that only covers what the prompt left open.
Inspect
Dextre reads the approved Azure scope, read-only, and reports what it found.
Generate
Terraform is filled from a tested template, and the pipeline is generated.
Approve
A named person approves the exact plan, with risky changes flagged.
Deploy
Only the approved plan deploys, to a development environment first.
Record
One change record goes to the audit trail and can be exported.
Capabilities
Ask once, ask only what is missing
A request that asks for what it needs and nothing else.
- The Azure deployment request is structured, and asks only for what the prompt leaves open.
- Approval is held while a cost, security or architecture decision is still open.
Look before it acts
Dextre reads the target environment first, and it cannot change it while reading.
- An administrator sets up the connection to an approved Azure scope. Dextre's own sign-in is read-only and separate from the pipeline's.
- An environment findings report lists what is found, missing or unknown. A missing required fact blocks the design approval.
- Tests cover revoked access, the wrong tenant and out-of-scope requests.
Infrastructure from tested templates
AI fills in a tested template. It does not write infrastructure freehand.
- Terraform is generated by filling a tested template, with every parameter checked and AI agents blocked from the infrastructure files.
- The generated Azure DevOps pipeline runs build, test, validation, preview, approval and deploy stages.
- The pipeline signs in to Azure without a stored password.
Approval by a named person
The approver sees the exact plan, and only that plan can deploy.
- The plan is shown at approval with risky changes flagged, and the approval is bound to that plan.
- Each approval records who approved by their Entra identity. The author of a run cannot approve it.
- A rerun cannot deploy twice, and the run page shows the deployment status.
Cost and secrets checked before go-live
The cost is on the table before anyone approves, and a missing secret stops the deployment.
- A cost estimate in Canadian dollars is shown before approval. The deployment is held when it exceeds the budget.
- A secrets checklist lists every Key Vault reference the app needs and holds the deployment on one that does not resolve.
Evidence, handover and recovery
The client keeps the record, and a failed run can pick up where it stopped.
- One cloud change record goes to the audit trail and can be exported in readable form.
- A retention period per client keeps a run past the default purge, with export to the client's own storage.
- A handover document in the delivery pull request covers architecture, resources, pipeline and operating notes.
- A run that fails part-way can be resumed. Development environments carry an expiry, and a person confirms any delete.
Existing applications, checked first
Dextre tells you what can move to Azure as it is, what needs a bridge, and what it will refuse.
- A deployability check gives each need of the app an Adapt, Bridge or Refuse result.
- It scans for AWS, Google Cloud and Firebase libraries and checks each dependency's health.
- A brownfield gate holds the codebase until its Azure connection, a user with sufficient rights and its pipelines are in place.
- A second Azure pattern, App Service with a database, deploys through the same flow.