AI & Agents
Agent Record API: Trace Runs, Tools, and Approvals
Follow the task from request to result with observable steps, scoped approvals, repeatable actions, and inspectable artifacts.
Read guide →🤖 RUNS / TOOLS / APPROVALS
Agent Record APIs.
Make every step and effect accountable.
Follow a task from a person’s request through observable steps, scoped approvals, retries, and delivered artifacts. Keep a record that can explain what happened without inventing hidden reasoning or collecting every private payload.
Separate runs from attempts
A run describes the user-level task. Steps and attempts explain retrieval, model work, tool calls, validation, and waiting. Preserve failures and retries instead of letting the latest attempt overwrite the earlier story.
Authorize before executing
A proposed tool call is not permission to act. Bind approval to the actual operation and arguments. Define reconciliation for uncertain external outcomes before enabling automatic retries that might duplicate a message or task.
Make observability fit the purpose
Operational traces and durable action records have different completeness and retention needs. Keep them correlated without relying on sampled telemetry as the only evidence of a required approval or external effect.
Read the complete guides.
Questions, answered.
Is a detailed agent transcript a complete audit?
Not automatically. Define the completeness, integrity, access, and retention requirements for the actions the application must account for.
Can a run always be replayed identically?
No such guarantee follows from recording alone. Sources, permissions, models, and external tools can change; treat replay as a new controlled attempt.