Testing & Debugging

Test Steps and full paths on the canvas before enabling live triggers. Draft is not a sandbox: clicking Run sends real messages, writes real records, and spends credits. Only Testing data, a pinned Step, and AI Agent Preview prevent external writes.

At a glanceVerified in app · Oct 30, 2026

Use it to: Test one Step, run from a specific Step, pin outputs, or preview writes before going Live
Find it: Step drawer → Test & Debug tab, or bottom dock → Run
Cost: Free for pinned Steps, Testing data, and writeMode preview · Live tests use Step credits
Limits: 256 KB UTF-8 JSON per Step Testing data · Polling sample fetch pulls up to 5 newest events (6/min)

At a glance

Mode	Where to start it	What runs	Writes	Credits
Testing data	Step → Test & Debug → Testing data	Saved sample stands in for the Step’s output.	No	No
Pin output for tests	Step → Test & Debug → Pin output	Test runs skip the Step and pass on Testing data.	No	No
AI Agent → Preview	AI Agent → Test & Debug → Preview	Real model and read-only tools; writes show `Proposed`.	No	Yes
AI Agent → Live test	AI Agent → Test & Debug → Live test	Agent only, with real tool calls and approvals.	Yes	Yes
Run workflow from here	Step → Test & Debug → Run from here	This Step and later Steps; earlier Steps reuse saved output.	Yes	Yes
Run (dock)	Bottom dock → Run	Full path from starting Step (Draft triggers use sample).	Yes	Yes
Live	Top status control → Live	Every live trigger event, schedule, and webhook call.	Yes	Yes

Draft and Live

Paused automatic triggersDraft Mode

Trigger card shows Draft mode. Active on manual Run.
Webhook URLs are paused and return 202 without starting a run.
Manual Run clicks execute real app actions and spend credits.

Armed automatic triggersLive Mode

Arms schedules, polling triggers, and public Webhook URLs.
Switching to Draft stops new events without undoing past runs.
Pinned Steps execute for real during Live trigger runs.

Test without touching real data

1. Seed Trigger Data2. Pin Write Steps3. Preview AI Agents4. Run & Verify App

Seed the trigger: Open Test & Debug → Testing data (or Load latest execution) with an edge case.
Pin write Steps: Save Testing data on write Steps and enable Pin output for tests.
Preview AI Agents: Run Preview first to inspect proposed recipients and payloads.
Verify the app: Point unpinned Steps at test destinations, click Run, and check the app.

What tests cost

Live tests and AI Agent Preview bill model and action credits under your credit guardrail. Testing data and pinned Steps cost 0 credits (Credits & Allowances).

Common surprises

Expectation	What happens	Fix
“Draft is safe to click Run”	Run sends real messages and writes records.	Pin write Steps or use test destinations.
Preview remembers the conversation	Preview starts fresh without reading or saving memory.	Use Live test to verify memory.
Webhook URL works in Draft	Draft pauses webhook URLs until Live.	Test with Testing data and Run, or switch to Live.
Run from step 5 replays steps 1 to 4	Steps 1 to 4 reuse their latest output or Testing data.	Start from the trigger to test the full path.
Approval-gated actions auto-pass in tests	Live test and Run pause in Executions.	Approve or Reject in Executions or Approvals.

Running automations and single steps

Skip a Step during testsPin Output for Tests

Freezes a single-execution action Step using saved Testing data (256 KB JSON).
Manual canvas and MCP test runs skip the Step at 0 credits and 0 ms.
Live runs, subflows, and single-Step tests still execute the Step for real.

Test the tail of a workflowRun from Here

Starts at the chosen action Step and runs everything connected after it.
Seeds up to 50 upstream Steps from latest output or saved Testing data.
Not available on Combine, await: all Steps, or Steps inside a Repeater.

Set a Run Point

Setting a run point

Glow sets the default run point on the trigger Step. Right-click a Step and select Set run point, or pick from Manual run starting step beside Run in the dock.

When Live at a deployed polling trigger, Run fetches live data. Starting at a Draft trigger uses its Testing data (or Load latest execution).

Run from Here

Running from a specific step (“Run from here”)

Right-click an action Step (or open ⋮) and click Run from here, or click Run workflow from here in Test & Debug.

Earlier Steps do not re-run. Glow seeds up to 50 referenced predecessors from their latest output or Testing data, reusing those event IDs. Unavailable on Combine, await: all Steps, or Steps inside a Repeater.

Step selected on the canvas. — Configure fields in Setup, inspect runs in Executions, and manage Testing data in Test & Debug.

On-canvas Setup Guide & AI repair

1. Connect your apps2. Run a quick test3. Publish when ready

Connect your apps: Connects the account, binds waiting Steps, and resumes parked Assistant builds.
Run a quick test: Runs from the starting point with Reconnect or Open step shortcuts.
Publish when ready: Switches the automation to Live once checks pass.

AI debug automation (“Use AI to fix”)

When an assistant-built or imported draft fails validation, click Use AI to fix in the Setup Guide to repair structural, semantic, and non-account issues while preserving disabled Steps.

Before going Live, review the Before you turn it on checklist. Track runs and spend in Activity and Billing & Usage.

Inspecting data with Executions

Use the Execution Log to find the first unexpected output, not just the final red Step:

Open the matching execution

Select the Step, open Executions, and expand the run matching your timestamp and execution ID.

Read the returned data or error

Inspect the JSON payload, click Copy output, or open Show full error.

Check the fields used downstream

Compare returned fields against downstream references to spot missing or mistyped keys.

Before you retry a write

If a write Step fails or times out, check the connected app first to avoid duplicate records or charges (Connections and External APIs).

Visual step statuses

Each Step updates its badge and outline to match the Execution Log and Step-Level Executions:

Status	Badge / colour	Meaning	Charged?
Running	Blue (`Execution in progress`)	Executing now.	Yes
Waiting	Blue (`Running…` / paused)	Paused on Wait, Human Review, or tool approval.	No wait charge; excluded from cap
Completed	Green (`Execution succeeded`)	Finished; output saved in Executions.	Yes
Failed	Red (`Execution failed`)	Stopped with error (partial loop failures show Amber `Execution issue found`).	Only work before failure
Skipped: not charged	Grey (`Skipped`)	Branch not taken or stopped upstream.	No
Stopped	Grey (`Stopped`)	Stopped manually.	Only work before stop

Execution counters

Step cards show lifetime run and retry counts. Clear them via Reset counters in the canvas utilities menu.

Debugging code and HTTP steps

Code editor: Use print() in a Code editor Step ({{ 4.result.executionOutput }}; valid JSON last line parses into result.output).
HTTP Request: Turn on Include Full Response to inspect status codes and headers alongside the body.

Verify it works

Run: Open Test & Debug on a Step, save sample JSON in Testing data, and enable Pin output for tests
You should see: A push-pin badge appears on the card, and manual Run tests use the saved sample at 0 credits.

What’s Next?

👉 Test an AI Agent →: Preview, Live test, and approvals in detail.
Execution Log: read a run Step by Step.
Error Handling & Retries: what happens when a Step fails.
Step-Level Executions: inspect input and output payloads for individual Steps.