Where do I see everything my agents did?#
Open Runs from the workspace tabs. The page lists every run in a table with its status, agent, trigger, start time, duration and cost, so you read a whole day of agent work in a few seconds. Each agent's own page in Agents also carries a Runs section, and the Agents dashboard shows Recent runs across your workspace.
| Control | Choices | Use it to |
|---|---|---|
| Status chips | All, Done, Errors, Running | Jump straight to the runs that need you |
| Agent picker | All agents, or any one agent | Audit one agent's week of work |
| Date range | Last 7 days, Last 30 days, This month, All time | Match a billing period or a client report |
| Runs per page | Page sizes from the dropdown | Scan long histories faster |
| Refresh and Clear filters | One click each | Pull the newest runs or reset the view |
What does a run receipt show?#
A run receipt shows what ran, in what order, what each step produced and what it cost. Click any row on the Runs page to open it. The header carries the agent's name, a status pill and an Open in Studio button that takes you to the canvas.
| Field | What it tells you |
|---|---|
| Trigger | What started the run: a test, a schedule, a webhook, a form or a manual press |
| Started and Finished | Exact local times, to the second |
| Elapsed | Total time from start to finish |
| Steps | How many nodes ran |
| Charged from credit | What the run drew from your prepaid credit |
| Estimated total | The sum of every step's estimated cost |
Below the summary, Estimated cost by step lists every node with its status and cost. A skipped step says skipped. A failed step carries its error text. Files an agent produced, such as a PDF report or a spreadsheet, sit under Generated outputs with a download button. A run that cost nothing says so in plain words.
| Label | Meaning | Your next move |
|---|---|---|
| Done | Every step on the main path finished | Read the output or download the files |
| Error | A step failed and the run stopped its branch | Open the receipt and read the Run failed box |
| Running | The run is still working | Watch it live in Studio |
| Didn't finish | The run stopped reporting before it ended | Press Run again |
Can I watch a run live?#
Yes. Press Run in Studio and the run dock streams each node's progress and log lines as they happen, then shows the outputs the moment the run ends. Runs on the durable engine add a Live monitor and controls panel on the receipt with queue wait, execution time, each step's attempt number and any retry the engine scheduled, so a flaky API call retries on its own instead of failing your customer.
What does an agent run cost?#
A run costs the sum of its cost-bearing steps, and the receipt itemizes every one. Logic steps such as Input, Output, Branch, Schedule, Webhook, Transform and Subflow cost nothing. Model steps cost according to the model you pick, and the LLM node shows each model tier's rate right in its settings. Agent Studio charges nothing for the HTTP Request step itself; the service you call bills you directly if it charges at all.
Live usage draws from your prepaid credit. Open Credit for your balance and Credit history, and see Pricing for plans and model rates.
How do spend caps stop a run from overspending?#
Agent Studio checks spend before every cost-bearing step, not after the bill arrives. Three guardrails work together:
- A per-run cost ceiling is checked before each paid node runs. If the next step would cross it, the run stops there and the receipt names the step.
- A per-agent daily cap reads durable run history before a run starts, so one busy agent never drains your credit overnight.
- Branch halting: when a step on the main path fails, the engine stops that branch instead of paying for downstream work that depended on it.
AI company employees add one more layer: a monthly cap per employee and per department that you set yourself. See Run a self-driving AI company.
How do I fix a failed run?#
Find, fix and rerun a failed step
Step 1: Filter to errors
Open Runs and press the Errors chip. Narrow it with the agent picker if you run many agents.Open RunsStep 2: Read the Run failed box
The receipt names the failing step in plain language. Open Technical details for the raw error when you need it.Step 3: Fix the step on the canvas
Press Open in Studio, adjust the node's settings or connection, and test the agent again for free.Open AgentsStep 4: Run it again
Back on the receipt, press Run again to replay the run with the same input.