Tracira icon

Tracira

Check and inspect your AI outputs in Tracira

Actions11

Output → Send an Output

AI-generated

Summary

Submit an AI-generated output with optional input context and attachments to Tracira for evaluation against configured rules, receiving a verdict, confidence score, and explanation; optionally wait for human approval or log asynchronously.

Inputs

  • Project Name (required) — Select or enter a project name; will auto-create the project on first use in Tracira.
  • Task Name — Optionally select or enter a task name within the project to categorize this output.
  • Input Text — Optional text representing the user's input or prompt given to the AI, shown as the user's side in the conversation.
  • Input Attachments — Optional files provided as AI input, which Tracira detects as images, audio, or documents; supports inline upload, URL, or files already stored in Tracira.
  • AI Output — The AI's generated output as plain text or JSON; optional if output attachments are provided.
  • Output Attachments — Optional files produced by the AI output (images, audio, documents) with the same attachment options as Input Attachments.
  • AI Model — Optional name of the AI model that generated the output, exactly as used (e.g., "gpt-4o").
  • After Check Mode — Determines workflow behavior after submission: wait for verdict synchronously (max 30s), log asynchronously for background evaluation, or hold for human approval before proceeding.
  • Action Name (required) — (When waiting for human approval) Machine name of the action the AI wants to perform, subject to human approval or rejection.
  • Action Summary (required) — (When waiting for approval) Plain-language description of what the AI's action entails, shown to the human reviewer.
  • Action Parameters (JSON) — (When waiting for approval) Optional JSON-formatted parameters relevant to the proposed action.
  • Callback URL — (When waiting for approval) Optional URL that Tracira calls upon human decision to resume the workflow; blank to use polling trigger.
  • Callback Events — (When waiting for approval) Specifies which events trigger the callback URL invocation.
  • Options — Optional metadata and contextual fields such as Actor ID, confidence score, cost in USD, instructions version, latency, output ID, revision linkage, session and subject IDs, and output timestamp override.

Output shape

A JSON object representing the AI output evaluation.

If 'After Check Mode' is 'verdict', the operation waits for and returns the evaluation including status, verdict, and confidence; if 'async', returns immediately with HTTP 202 and logs evaluation in background; if 'approval', returns immediately holding the output for human review with action details; attachments and metadata are included if provided.

Examples

Example 1: Sending an AI-generated message for evaluation

Specify the project and task names, provide the AI output text, optionally include the input prompt and attachments, set the mode to wait for the verdict to branch based on evaluation results.

Links

Discussion