What does a dry run test?#
A dry run is a complete test of your agent's structure. Steps that only compute run for real. Steps that would spend money or reach the outside world return a simulated result instead, so you see the full shape of a run without sending a text, charging credit or touching a customer.
| Step type | In a dry run | Examples |
|---|---|---|
| Logic | Runs for real | Branch, Switch, Transform, Aggregate, Loop, Subflow, Wait |
| Documents and data | Runs for real | Extract PDF Text, Parse Spreadsheet, Filter and Clean Rows, Generate CSV, Knowledge Search |
| Triggers and I/O | Runs for real with your default fields | Schedule, Webhook, CRM Trigger, Input, Output |
| AI | Simulated result | LLM (Claude), Classify, Extract Fields, Transcribe Audio |
| Outside apps and messages | Simulated result | HTTP Request, Fetch a URL, Send Text, Send Email, Slack Message, CRM updates |
| SEO and search data | Simulated result | Keyword Research, Site Audit, Rank Tracker, AI Visibility Tracker |
- $0for every dry-run test, however many you run
- 26node types that run for real in a dry run
- 96node types that return a simulated result instead of spending or sending
How do I run a free test?#
Test an agent without spending
Step 1: Open the agent on the canvas
Open any agent from your dashboard, or open a template to practice on.Open your agentsStep 2: Press Run test
The button sits in the canvas header. The run dock opens and every step streams its status and output as it finishes.Step 3: Read each step
Select a step in the dock to see what it received and what it returned. A failed step shows its error, and steps downstream of it never run against bad data.Step 4: Test one step on its own
Right-click a card and choose Run this step to test a single piece. Agent Studio asks for the values that would have arrived from upstream.Step 5: Fix and run again
Change a field, rewire a branch, rerun. Dry runs are free, so test as often as you like.
When should I run it for real?#
Once the dry run passes. A dry run proves the structure: wiring, branch logic, transform expressions, loop behavior and output shape. It does not judge prompt quality or how an outside service answers. For that, turn on Run for real · uses credit in the run dock and run a handful of live passes against real inputs. AI steps call the model, web requests go out and paid tools run, all billed from your prepaid credit.
| Run test (dry run) | Run for real | |
|---|---|---|
| Cost | Free | Uses credit at live rates (pricing) |
| AI steps | Simulated output | Real model output |
| Messages and outside apps | Simulated, nothing sent | Sent and called for real |
| What it proves | Wiring, branches, loops, output shape | Prompt quality and real-world answers |
| Spend guard | Nothing to guard | A per-run cost ceiling checked before every paid step |
Do loops and subflows run in a dry run?#
Yes. A Loop genuinely iterates over your list, and a Subflow runs its inner flow, both in dry-run mode all the way down. That means you can test fan-out over 50 rows or a nested workflow and see every iteration, with no chance of anything inside spending credit.