AI runtime profiles

Separate intent from resources

A Flow describes the work. A runtime profile describes applicable resources and constraints: AI capability, connection, scope and limits. This lets an app be reused without imposing its author’s provider accounts on everyone.

Parameter	Expected values	Description
Capability	Function needed for work	Text, analysis or another admitted capability
Connection	Configured, authorized access	Usable provider
Resources	Provider-supported choices	Execution parameters
Residency	Effective region and constraints	Processing boundary
Limits	Budget and usage policy	Bounds execution

Example

Quote preparation uses Sales drafting. The profile selects an authorized connection, the organization’s required region and usage limits. Users see why a resource is unavailable and which setting needs checking.

Refuse incompatible execution

Missing capabilities, revoked connections and incompatible regions must prevent the affected execution. Profiles do not bypass permissions or switch to unauthorized providers or regions to finish work.

See Data residency and Usage and limits to prepare team configuration.

Configure your resources

From account settings, open LLM Runtime, then choose the account and application to configure. Available locations match the responsibilities declared by that application; no connection or profile is chosen for you.

1. Add a Connection and choose its provider and authentication method. Enter a personal credential only in the secure account portal, never in chat or a Surface.
2. Prepare a Profile with an exact connection, an authorized model and your input, output, time and cost ceilings. Saving does not validate the provider.
3. Validate the connection for the chosen Profile and model. When a linked configuration changes, explicitly select its new revision before using it.
4. Write the Worker instructions and link the intended Profile. Account, personal usage, Profile and Worker budgets jointly bound execution; they do not grant additional access.

A lost save response keeps the original request available for recovery. Do not create a second configuration to bypass a pending result; check the original operation first.

A new or edited text Profile is saved as setup required, even if its Connection was previously ready. Validate that exact text Profile again, then explicitly select its new revision for the intended Worker. Other Profiles and Workers keep their original revisions; Surface does not repair shared references or restart AI automatically.

Replacing a credential requires explicit confirmation. The previous credential is never displayed again. Validate the connection afterwards and select its updated revision in linked Profiles. If the replacement response is missing, check the same operation’s result instead of entering and sending the credential again.

Review a text proposal before saving

Choose the Task, Worker and exact document, Anchor, text Field or selection to assist. Review the proposed plan, sources and limits, then explicitly confirm that same plan. Surface uses the selected Profile’s validated text capability; it does not choose another provider or turn a structured Worker result into text.

Live text is only a temporary preview. The final proposal has its own retained result and receipt; accepting it prepares your ordinary editable draft, not a completed target save or Task. Keep your existing draft when a conflict appears. This strict consultation requires the same canonical target revision; a changed revision is not silently rebased. A missing response is recovered from the original operation rather than generating again.

Cancellation requests stop further authorized work but do not promise that a provider has already stopped or that reserved usage has been refunded. Check the actual execution state.

Resume a delegated question explicitly

When a Worker asks for information, its Execution waits without completing the Task or recording an intermediate failed Result. If the question expires, the same Execution and Task remain waiting for an explicit resumption; expiry never launches another AI request automatically.

Review the question and confirm its resumption. Surface checks your current permissions, the original execution deadline and the remaining budgets again. Resuming does not reset these limits. If the original execution deadline has passed, a new explicit plan is required; the old reservation and operation cannot be revived silently. After a missing response, recover the original operation before submitting another answer.

Applied examples : Distinguish context and resources

The policy defines AI sources and the Flow bounds work. Select the runtime profile in Surface settings.

Configuration

Fragment placed in contextPolicies[]. Properties and accepted values.

{
  "key": "quote-context",
  "targetDefinitionPaths": [
    "workspace/quotes/surface/quote"
  ],
  "sources": [
    {
      "key": "client",
      "kind": "relation",
      "definitionPaths": [
        "workspace/clients/surface/client"
      ],
      "linkTypePaths": [
        "application/client"
      ],
      "queryKey": null,
      "required": true,
      "revision": "current",
      "priority": 1,
      "maximumItems": 1,
      "maximumCharacters": 8000
    }
  ],
  "maximumCharacters": 16000
}

Variation and connection

Fragment placed in flowBlueprints[].composition.budgets. Properties and accepted values.

{
  "maximumDepth": 2,
  "maximumOccurrences": 20,
  "maximumConcurrency": 2,
  "maximumIterations": 3,
  "maximumDurationMilliseconds": 86400000,
  "maximumCostUnits": 100,
  "maximumStateBytes": 32000
}

UI layout example

The illustrative layout below places this configuration in the Clients/Quotes app. Insert each fragment at the stated location, with Field, Query and Action references matching the Model declarations.

1. Runtime profile
2. Connections
3. Residency
4. Save

SurfaceClients / North StudioExample

AI runtime profilesDistinguish context and resources

Clients

Runtime profile

Team profile

Required sources

Client and pricing

Flow budget

100