The Tester is the panel that opens from the canvas to test a workflow before publishing it. You can send messages as if you were an end user, see every node that executes, and inspect input, output, and state at any step of the flow. It shares the same visual interface as the production Debugger, so learning one is learning the other.Documentation Index
Fetch the complete documentation index at: https://docs.jelou.ai/llms.txt
Use this file to discover all available pages before exploring further.
Open the Tester
In the canvas top bar, click Test. A side panel will open with the workflow header, the conversation area, and a field to type messages.If the workflow has never been executed, the first message you send starts a new session. Each session counts as one execution (see Workflow Executions).
Chat view and Execution view
The panel has a toggle at the bottom with two modes:| Mode | What it shows |
|---|---|
| Chat | Only the conversation between the bot and the simulated user. Useful to validate the conversational flow without distractions. |
| Execution | Conversation + indicators for every workflow node that executed (icon, name, and status). Useful to debug internal logic. |
Node indicators
In Execution view, every workflow node that runs during the test appears as a small card interleaved with the messages. The card shows:- Canvas icon of the node (Start, HTTP, Code, AI Agent, Conditional, Tool, etc.)
- Node name as configured in the canvas
- Status (green for success, red for error)
N nodes · Xms total). If the workflow calls a sub-skill, the inner sub-skill nodes appear as a separate group to mark the execution boundary.
Node detail panel
Clicking on any indicator opens a resizable side panel with the full execution information for that node:Input
Input
The exact payload that entered the node.
Output
Output
What the node returned (response, generated message, conditional result, etc.).
Initial state / Final state
Initial state / Final state
Snapshot of
$memory and $context before and after the node executed.Errors
Errors
If the node failed, message and stack trace.
Copy the path of a value
Click any key inside the JSON ofInitial state or Final state to copy the full path in bracket notation to the clipboard. For example, clicking the url key inside finalState.tool copies finalState.tool.url. A “Copied ✓” tooltip confirms the action.
Detail panel actions
Above the input/output, the panel has contextual buttons:Go to node
Centers the canvas behind the Tester on the corresponding node. Only appears for nodes in the active workflow. For internal sub-skill nodes, the button is hidden because the node does not live in this canvas.
Interactive bubbles
Bubbles that are interactive in production (buttons, lists, CTAs) are now clickable inside the Tester. You don’t have to manually type the response to advance the flow: click the button as the end user would.| Bubble type | How to simulate |
|---|---|
| Quick replies | Pill-style buttons under the message. Click sends the payload.targetId and advances to the destination node. |
| List options | Cards with title and optional description. Click resolves the option against node.configuration.messages[0].options and advances through the correct branch. |
| CTA URL | Button with an external-link icon. Click opens the URL in a new browser tab. Does not advance the flow (same as on WhatsApp). |
| Carousel | Horizontal cards with image, title, and description. The URL button opens the link in a new tab. |
Use cases
Debug an API failing in production
Debug an API failing in production
A telco team builds a flow that validates invoices via API. The API responds inconsistently and the flow breaks in production. They run the test in the Tester, click the failing HTTP node, see the full request and response, and copy the path of the problematic field to report it to the backend team, all without leaving the builder.
Validate Conditional logic
Validate Conditional logic
A retail QA team validates subscription billing flows before each release. One Conditional branch always causes problems because the matching logic depends on strings with accents. They run the test, see exactly which term matched in the Conditional node panel, and fix it without escalating to engineering.
Validate an AI Agent response
Validate an AI Agent response
A product manager configures an AI Agent that asks for name and email and saves them into
$memory.datosUsuarios. From the Tester they confirm the Final state has the correct structure ({ nombre, correo }), copy the path finalState.datosUsuarios.correo, and use it in the next text node without rebuilding it manually.Test a sub-skill from the parent flow
Test a sub-skill from the parent flow
A parent workflow calls an identity-validation sub-skill. In the Tester they see the sub-skill rendered as a separate group of indicators. They click the Tool node inside the sub-skill and use Debug Tool to inspect the Tool’s internal execution without opening the sub-skill in another tab.
Simulate a conversation with buttons
Simulate a conversation with buttons
Limitations
- Inner sub-skill nodes: the Go to node button does not apply to nodes that live on a different canvas than the active one. To inspect them, open the sub-skill separately or use Debug Tool if the container is a Tool.
- Tests don’t affect production: the Tester session is independent of real traffic. Conversations, variables, and memory you generate here are not logged in the production Debugger and don’t count as real user conversations (they do count as executions for billing).
- One session at a time: if you close the Tester without finishing a conversation, the session is discarded. The next time you click Test, a new one starts.
Related articles
Your first workflow
Build your first workflow step by step and test it from the Tester.
Workflow executions
How executions are counted and their impact on billing.
Share a test link
Generate a public link or private WhatsApp session so your team can test the workflow.
Publish versions
When the flow is ready in the Tester, publish a version for production.