# Review capabilities and deployment

> Move from requested capabilities to approved access, a verified deployment and a controlled budget.

Source: https://diakrin.com/docs/governance

Diakrin lets an app’s owner and IT team review the services, permissions, costs and deployment requirements a build needs before it reaches employees. Always use the state shown for the exact app and environment.

## Read capability states

A capability moves through separate states, and only the last one means the app can actually use it:

```steps
Requested | Builder or agent | The app says what it wants, such as reading a named data source.
Approved | Policy or IT | Company policy or an administrator permits it, with limits. The app and each employee need their own grants.
Configured | Authorized owner | The connection is set up, for example credentials entered on the customer-hosted form.
Verified | Your cloud | The connection is checked in your environment and the app can use the capability.
```

A capability can also be **unavailable**, **blocked** or **waiting for approval**. Ask your administrator to resolve it instead of working around a blocked action.

Open the app’s capabilities and policies, check each relevant permission and ask your administrator to resolve blocked or missing access. Selecting a capability in the app brief does not approve it.

## Review before deployment

| Check | Confirm with your owner or IT team |
| --- | --- |
| Environment | The exact test, staging or production target |
| Audience | Who should open the deployed application |
| Data | Approved sources, operations and record-level access |
| Runtime | A connected, compatible company-cloud runtime |
| Cost | Applicable limits, estimates and retained costs |
| Approval | The person authorized to accept the plan |
| Recovery | How a failed release is diagnosed or rolled back |

If the console says a plan, approval or deployment action is unavailable, complete its prerequisite with your onboarding contact.

## When a build, deployment or publish is refused

Every build, deployment and publish is checked again at the moment it would change your company cloud. A refusal changes nothing in the cloud. The error your coding agent or CLI receives says which rule stopped it and what to do next.

| You see | What it means | What to do |
| --- | --- | --- |
| A policy id such as `IAM-02` or `COST-01` | The company rule that stopped the action | Read the repair text next to it |
| Verdict `repair` | You can change the request | Apply the repair, then try again |
| Verdict `escalate` | Someone else has to act, for example raise a budget or refresh the security scanner | Send IT the policy id and the `intentId` |
| An `intentId` | The record IT uses to find the refusal in the audit log | Include it when you ask for help |

Common refusals: no approved budget for this project (`COST-01`), the security scanner is out of date (`VULN-03`), your role does not allow the action (`GOV-04`), or audit records cannot be stored right now (`GOV-06`).

Budget refusals, and what the owner does:

| Code | What it means | What the owner does |
| --- | --- | --- |
| `BUDGET_REQUIRED` | No owner-approved build, runtime and total budget covers this app | Approve **Organization spending limits** in **Finish setup**, on the deployment prerequisites step |
| `BUDGET_EXHAUSTED` | This month's approved limit is used up | Raise the limit, or wait for the next month or for held amounts to settle |

Nothing is built or deployed on a refusal. Each build or deploy holds budget for that one operation; an operation whose outcome is unknown keeps its hold until it is reconciled.

## Confirm the application is usable

A deployment is ready for team use only after its actual URL works for an authorized user, unauthorized access is denied, the intended workflow succeeds and its owner confirms the result. Record the release/environment you tested.

Project creation, local tests, preview data, a successful build and a deployment status are separate evidence. Check the real app before announcing it to your team.

## Operate the app

Keep ownership, access and spending responsibility current. Before updates or rollback, review the affected environment and release. Pause, retirement and ownership actions require their own deployed capability and permission checks; contact IT when the console reports them as blocked.

> [!SELF-HOSTED]
> On your AWS installation: app errors come from the app's own CloudWatch log group, at most one hour and 100 rows per read, codes and IDs only, never message text. Ask your agent (`app.logs.read`) or run `diakrin app observations read` (see [When your app fails](/docs/troubleshooting#when-your-app-fails)). Request counts for each app are on **Govern → Adoption** (see [App requests on the Adoption page](/docs/dashboard#app-requests-on-the-adoption-page)). The app data inventory shows **Unavailable** for now, because no data collector runs in self-hosted AWS yet; that means unknown, not empty.

For console and connection failures, use [Troubleshooting and help](/docs/troubleshooting).

## App budget and cost estimate

Every app has a monthly budget ceiling. Deployments that would exceed it are refused at admission, even when the app's project still has budget left. Company admins and project managers can change the ceiling and the alert thresholds (80 % and 100 % by default) from the app's **Cost** card.

The cost shown is an estimate. Diakrin observes each app's usage (container size and running hours, requests, data transfer, log volume, build minutes and storage) and multiplies it by the cloud provider's public on-demand prices, plus a share of the setup's shared baseline weighted by compute. It is approximate. It is not your cloud bill and does not include discounts, savings plans, credits or taxes. Where Diakrin cannot read a metric, that part is listed as unknown and the read is marked partial; it is never counted as zero or guessed.

## Costs and budgets you can trust

Every estimate is priced from a named price source with an as-of time and a region, and the **Cost** card shows them. Each read states whether its pricing is **priced**, **partial**, **stale** or **unavailable**; when prices are stale or missing the amount stays empty rather than becoming a number you cannot defend. Components Diakrin could not price are listed by day, environment and component, and are never shown as zero. The setup's shared baseline (the infrastructure every app in a setup uses) is split between apps by their compute share and counted once, so totals across apps never double-count it.

## Shared project policy

Every app belongs to exactly one Project (an automatic Default exists in every organization) and always has its own authority scope: apps never share permissions or deployment policy. Apps split from a former shared group keep spending against that group's existing ceilings, so a split never removes a limit. A Project carries inherited settings that can only make its apps stricter, unless IT delegates a bounded choice. Moving an app between Projects is an IT action; it never changes the app's own access or scope.
