Agent observability See what the agent could do.
See what the agent could do.
See what it actually did.
Tenure captures tools, calls, arguments, results, and evidence paths directly on the model request path. Start with the behavior your agent actually produces, not the behavior a dashboard guesses after the fact.
Observation shows what happened. Reviewed policy decides what should remain possible.
The complete trajectory 01Available toolsEverything the application made visible to the model. → 02Tool invocationThe selected tool, call ID, arguments, and workflow ID. → 03Tool resultThe matching output, status, and original call ID. → 04Next actionThe later call that used the earlier result as evidence.
A log of the answer is not a log of the agent.
An agent can expose ten tools, invoke three, pass evidence between two, and leave one result pending. Tenure records each part of that sequence while it crosses the wire.
Exposure and use Available Invoked
Get insight into available tools and invoked tools.
What could the model choose?
fetch_customerEXPOSEDgenerate_emailEXPOSEDsend_emailEXPOSEDdelete_customerUNUSEDAvailable tools reveal the total capability surface presented to the model, including tools it never selected.
What did the model attempt?
fetch_customerCALLEDgenerate_emailCALLEDsend_emailCALLEDInvoked tools show the actions selected during a run, with their inputs, outputs, and relationship to the surrounding workflow.
A tool that was available but unused is still part of the agent's exposed authority. Tenure keeps that distinction visible when you review policy.
Workflow reconstruction CALL → RESULT → CALL → RESULT Identity Continuity Evidence
Follow the evidence from one action to the next.
When your application returns tool results, Tenure associates later actions with the results that preceded them.
fetch_customercustomer_id: cus_9482active: truetier: enterprisegenerate_emailcustomer_id: cus_9482draft_ready: truereview requiredAttribute every run
A separate agent token ties activity to the correct workload and active policy.
Preserve call relationships
Call IDs and workflow IDs connect invocations with their returned results.
Inspect what supported the action
See earlier arguments and result fields used by downstream actions.
From observation to policy 01ObserveCapture representative workflows. → 02DraftPropose tools and preconditions. → 03ReviewRemove unnecessary authority. → 04PublishBind policy to agent identity. → 05EnforceCheck every governed request.
Turn real behavior into a reviewable boundary.
Tenure uses successful observed workflows to produce a draft. Your team decides which capabilities and evidence requirements become enforceable policy.
Built to be inspected Request record Observation record Policy record
Keep the request, workflow, and decision connected.
What entered the model call?
Inspect protocol, provider, model, route, request ID, available tools, and exposed tools.
What happened during the turn?
Inspect released calls, arguments, results, status, call IDs, and workflow relationships.
What changed after review?
See the allowed capabilities, required evidence, active version, and enforcement outcome.
Arguments and results may contain sensitive application data. Tenure stores observation payloads encrypted and returns them through the authorized agent detail view.
Start with evidence
Observe the agent you actually run.
Route supported OpenAI or Anthropic traffic through Tenure, run representative workflows, and turn what you observe into policy.