Observability and Cost Controls

Monitor agent runs, trace tool activity, review cost, and use evaluation gates for safer autonomous workflows.

Observability surfaces help you understand what autonomous agents are doing, how much they cost, and where a workflow failed. Use them when running multi-step tasks, scheduled automations, channel bots, or autonomous pipelines.

Dashboard Overview

Open the dashboard to review:

Running, completed, failed, and stopped task counts
Recent agent activity
Task flow over time
Cost summary
Operational signals for autonomous work

The dashboard is the fastest place to check whether the system is healthy before starting more work.

Task Trace

Task traces show execution details for a single run. Depending on the task, you may see:

Planning steps
Tool calls
Runtime events
Errors and retries
Generated artifacts
Cost and latency details

Use traces to answer "what did the agent do?" before deciding whether to trust, rerun, or revise the output.

Cost Panel

Cost controls help you understand spend across providers, models, and tasks. Review cost when:

A task runs longer than expected
A model choice changes
You enable higher reasoning
Automations run on a schedule
Channel bots are available to multiple users

If cost is unexpectedly high, lower reasoning, narrow the prompt, reduce automation frequency, or use a cheaper model for routine work.

Evaluation Gates

Evaluation gates are checks that can run before a workflow is considered ready. They are useful for autonomous pipelines and release-like flows.

Use gates for:

Required tests
Lint or typecheck checks
Review policies
Cost or budget limits
Confidence thresholds

If a gate fails, inspect the failure before asking the agent to continue.

Operational Workflow

Start with the dashboard to confirm system health
Open the task trace for a specific run
Review tool activity and generated artifacts
Check cost and model usage
Approve, rerun, or stop the workflow based on the evidence

For recurring work, combine observability with automations and approvals so the agent can run independently while still producing reviewable evidence.

Cost Panel

Cost controls help you understand spend across providers, models, and tasks. Review cost when:

A task runs longer than expected

A model choice changes

You enable higher reasoning

Automations run on a schedule

Channel bots are available to multiple users

If cost is unexpectedly high, lower reasoning, narrow the prompt, reduce automation frequency, or use a cheaper model for routine work.

Evaluation Gates

Evaluation gates are checks that can run before a workflow is considered ready. They are useful for autonomous pipelines and release-like flows.

Use gates for:

Required tests

Lint or typecheck checks

Review policies

Cost or budget limits

Confidence thresholds

If a gate fails, inspect the failure before asking the agent to continue.

Operational Workflow

Start with the dashboard to confirm system health

Open the task trace for a specific run

Review tool activity and generated artifacts

Check cost and model usage

Approve, rerun, or stop the workflow based on the evidence

For recurring work, combine observability with automations and approvals so the agent can run independently while still producing reviewable evidence.