Standard
  • Docs
  • Examples
  • Pricing
  • About

Get started

  • Overview
  • Conversation analytics
  • How Standard works
  • Run your first task
  • Connect
  • Join a Space

Build

  • Authoring · Sales calculation
  • Examples
  • Software release verification
  • Operation profiles
  • Integration reference

Concepts

  • Sources and customer vaults
  • Inputs and permissions
  • Accounts and billing
  • Results and lifecycle
  • Verification checks
  • Follow-on tasks
  • Cleanup

Security

  • Security model
  • Threat model and AI risk mapping
Documentation menu

Get started

  • Overview
  • Conversation analytics
  • How Standard works
  • Run your first task
  • Connect
  • Join a Space

Build

  • Authoring · Sales calculation
  • Examples
  • Software release verification
  • Operation profiles
  • Integration reference

Concepts

  • Sources and customer vaults
  • Inputs and permissions
  • Accounts and billing
  • Results and lifecycle
  • Verification checks
  • Follow-on tasks
  • Cleanup

Security

  • Security model
  • Threat model and AI risk mapping
Standard

Managed confidential execution for approved functions.

© 2026 Standard Runtime, Inc. All rights reserved.

  • Security
  • Docs
  • Examples
  • Pricing
  • FAQ
  • About
  • Contact
  • Terms
  • Privacy
  • Cookies
  • Docs
  • Examples
  • Pricing
  • About
  • Get started

    • Overview
    • Conversation analytics
    • How Standard works
    • Run your first task
    • Connect
    • Join a Space

    Build

    • Authoring · Sales calculation
    • Examples
    • Software release verification
    • Operation profiles
    • Integration reference

    Concepts

    • Sources and customer vaults
    • Inputs and permissions
    • Accounts and billing
    • Results and lifecycle
    • Verification checks
    • Follow-on tasks
    • Cleanup

    Security

    • Security model
    • Threat model and AI risk mapping

    Get started

    • Overview
    • Conversation analytics
    • How Standard works
    • Run your first task
    • Connect
    • Join a Space

    Build

    • Authoring · Sales calculation
    • Examples
    • Software release verification
    • Operation profiles
    • Integration reference

    Concepts

    • Sources and customer vaults
    • Inputs and permissions
    • Accounts and billing
    • Results and lifecycle
    • Verification checks
    • Follow-on tasks
    • Cleanup

    Security

    • Security model
    • Threat model and AI risk mapping

    Run your first task

    Follow Conversation analytics from an approved function and conversation batch to verified counts and a separately authorized report update.

    This walkthrough follows Conversation analytics: the Analytics agent requests conversation-categories for batch-042, and the Analytics integration receives permitted category counts. The input owner, agent and receiving integration can belong to one organization while holding distinct responsibilities.

    The figures and operation sequence are illustrative. The downloadable classifier is JavaScript reference logic, not a deployed Wasm function or a complete install-and-run project. The steps below describe the setup to agree before using your own accepted workload; they do not imply that conversation-categories is already installed in your account.

    Agree the function and setup

    Request access and complete the assisted setup checklist. Agree the supported execution route, input format, limits, recipient and verification references. An installed owner host, a hosted requesting agent and an authorized recipient have different responsibilities.

    Current hosted limit: connecting and approving a hosted request does not establish that it can execute. When admission is approved but Control has reported no progress, the hosted status route returns hosted_continuation_unavailable; the continuation transport is not available on that path. Repeated polling does not start it. Confirm the supported execution path during setup. See Connection and execution availability.

    1. Create the Space

    The sponsoring organization creates the reusable Space and pays for the execution it sponsors. Define the input owner, Analytics agent and Analytics integration. Sponsoring or signing in grants no execution or release permission. See Accounts, Spaces and billing.

    2. Deploy the function and its profile

    Use the conversation classifier reference to define category rules, the input format and duplicate-ID rejection. Adapt that logic to the supported Wasm interface and prepare a complete project, operation profile and fixture suite. Test the exact artifact before publication. Any model used by the classifier must run inside the protected runtime.

    The permitted output contract includes category counts and redacted failure reports. A duplicate conversation ID produces a failure report with no counts. The profile also names the input slots, recipient and execution limits; it does not authorize use of the input.

    Write a protected function explains the interface and real build/test/deploy commands using a separate sales-calculation example. Those example files and binary formats are not the conversation classifier. During setup, supply the corresponding classifier project, profile and fixtures before using the commands for .

    PreviousHow Standard worksNextConnect

    On this page

    • Agree the function and setup
    • 1. Create the Space
    • 2. Deploy the function and its profile
    • 3. Authorize
    • 4. Run
    • 5. Receive and verify the result
    • Handle refusals and failures
    • Add owners when the function needs them
    conversation-categories

    3. Authorize

    The input owner approves the selected batch, classifier revision and release of the permitted result to the Analytics integration. For batch-042, bind the approved connection, resource selection, version and required digest and size checks. Connected-source work needs per-operation authority.

    The Analytics agent requests the function using references. Raw conversations and source credentials are not supplied to its environment. Connection verification and approval-time snapshots can precede final operation approval under separate scoped permissions. See Sources and customer vaults.

    4. Run

    The requested operation is: “Classify batch-042 and return category counts.” The Analytics agent requests it using the identifiers and fields its MCP tool schema supplies. The accepted profile, source selection, recipient and permissions must already be configured; the example names alone grant nothing.

    On a configured execution path, confidential ingress retrieves and validates the batch and the approved classifier processes it inside the confidential runtime. The integration reference describes the implemented local-file and connected-source request contracts and their custody restrictions.

    5. Receive and verify the result

    The hosted response contains operation and delivery metadata. The Analytics integration retrieves the permitted result and verifies Standard Proof against trusted function, input, authority, runtime, recipient and result references before making the counts available to the Analytics agent.

    For an installed recipient, the SDK returns verified result bytes as value, plus a proof verification summary. It does not return a report property. Decode and validate the bytes against the classifier’s accepted output contract. Hosted custody returns delivery metadata instead. See Receiving results in the SDK for the actual return shapes.

    The illustrative result is Coding 420, Writing 310, Research 190 and Other 80: 1,000 conversations in total. The reporting workflow separately decides whether publication is permitted.

    Valid evidence can still report failed checks. A duplicate-ID failure report verifies the failed operation and contains no category counts. Verified counts can also remain held for publication approval; invalid evidence blocks acceptance entirely. See Verification checks.

    Handle refusals and failures

    OutcomeApplication behavior
    Missing or expired authorityKeep the operation blocked; obtain the required authority or narrow the request
    Unsupported input or requested limits exceededCorrect the format or resource request; do not treat the refusal as a result
    A limit is reached during executionInspect the reported failure and stage; do not accept a partial run as success
    Release refusedNo protected result reaches that recipient; successful execution does not override the refusal
    Required verification failsReject the result and stop the dependent action
    Timeout or uncertain statusReconcile the existing operation before resubmitting; follow its bounded retry and idempotency rules

    Execution, delivery and cleanup are separate statuses. Retain the operation identifier and safely redacted failure information for diagnosis. See Results and lifecycle.

    Add owners when the function needs them

    The same process can involve a vendor’s program or a partner’s records. Each required owner authorizes its own contribution and result release. An invitation, account connection or payment grants no authority over another owner’s assets. See Join a Space.