Getting to know Studio
Agent Studio is where you build, test, and activate agents. It lives inside Imaginne, at ai.nnumbers.com.br/agent, and uses the same session: there is no separate login.
Three ways to build the same agent
| Way | What it looks like | When to use |
|---|---|---|
| Conversation | You describe it, Imaginne builds it | Starting from scratch, changing without hunting for where it is, understanding a failure |
| Simple mode | A numbered list of steps, connected automatically | Most agents: one step after another |
| Visual editor | A canvas with boxes and connections | When the flow has parallel paths, branches, or joins |
All three act on the same document. What the conversation creates appears in the list and on the canvas immediately; what you move with the mouse is what the conversation sees in your next sentence. See Building by conversation.
Simple mode is the default for manual editing. When a flow comes to have parallel branches, it refuses to show it as a list and takes you to the visual editor — changing nothing. A list that hid a branch would show a flow that is not the one that runs.
Finding things
| Feature | How to get there |
|---|---|
| Command palette | Cmd/Ctrl + K — jump to any agent, space or action without navigating |
| Explore | the gallery of ready-made examples, by category, with a button that really creates one |
| Favourites | star an agent and it stays within reach; the mark is yours and follows your user |
| Home | the entry points, your recent agents and the organization's pulse |
The component library
When you add a step you choose from a searchable library — it matches synonyms and ignores accents —, with suggestions and the ones you used most recently.
You never type a tool's technical name anywhere. See System components.
Starting from something that already exists
| Path | What it does |
|---|---|
| Import cURL | paste a curl and the HTTP step is born configured; a credential in the command becomes a secret or is left out |
| Import OpenAPI | point at a 3.x document, pick the operation, and the step is assembled |
| Capture from the webhook | fire a real call and the captured shape becomes the input schema |
| Save as template | turn a finished agent into a template, yours or the organization's, marking what becomes a question |
The working cycle
Save → Test → Activate
| Action | What it does |
|---|---|
| Save | Stores the draft. Affects nothing that is running |
| Test | Runs it for real, with data you provide |
| Activate | Publishes an immutable version and makes it active in an environment |
Testing is a real run. It is not a simulation: it creates a durable run, with a timeline, and goes through the same permissions as any other. That is why it can be used to decide whether to activate.
The step panel
Clicking a step opens its configuration panel on the right. It is organized into fixed blocks:
| Block | Content |
|---|---|
| Configuration | The essentials of that step |
| Input data | What it receives from earlier steps |
| Expected output | What it should return |
| If it fails | What happens when it does not work |
| Advanced | Timeout, retries, and fine tuning |
The panel can be resized and opened full screen, and the width you choose is remembered.
The step summary
Each step shows, in one line, what it receives · what it does · what it produces. That is what lets you read a ten-step flow without opening ten panels.
The "Data" button
You never type a data item's technical path. Where a field accepts information from another step, the Data button opens a picker with what is available at that point in the flow — and nothing else.
Inside a loop, "the item" and "the index" appear in the picker; outside it, they do not, because there they do not exist.
A successful step test pins its result automatically, and from then on the picker offers the real fields of that response — including nested ones, like body.customer.address.city. Renaming a step does not break references: they point at a stable identifier, not at the name.
See Data between steps.
Available resources
Tools, skills, and connections appear with each one's real state:
| State | Meaning |
|---|---|
| Available | It can be used now |
| Not allowed | It exists, and the space has not allowed it |
| Unavailable | It does not exist in this installation |
| Not confirmed | It could not be checked right now |
"Not confirmed" never becomes "does not exist": the two would lead to opposite decisions. There is a button to check again.
See Allowed resources.
The status bar
At the top of the editor, Studio shows where the agent stands: whether there is an unpublished draft, which version is active in each environment, and what is missing to activate. That is where warnings like "this change has not been published yet" come from.
The theme follows yours
Light or dark, the choice is yours and it is the same one as WebChat: changing it in one applies to the other.
Next steps
Was this page helpful?
Report a problem on this pageDo not send passwords, keys, tokens, or customer data.