> ## Documentation Index
> Fetch the complete documentation index at: https://vida.io/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Reusable helpers and browser automation

> Turn a stable software operation into a tested function a Vida Agent can call.

A reusable helper gives a Computer Agent one exact operation with defined inputs and a structured result. Use a helper when the same work should run consistently without asking the Agent to plan the procedure again each time.

Helpers can be:

* **Browser-backed:** Use the Computer Agent's signed-in browser to navigate a website or call an application endpoint that depends on the browser session.
* **Browser-independent:** Work directly with an API, file, or data source without opening the browser.

## Choose a helper or delegated work

| Need                                                     | Best choice                |
| -------------------------------------------------------- | -------------------------- |
| One stable operation with known inputs and results       | Reusable helper            |
| A prepared workflow in a signed-in website               | Browser-backed helper      |
| API, file, or data work with no browser dependency       | Browser-independent helper |
| Novel work that requires planning across several systems | Computer delegation        |
| Interactive collaboration and clarification              | Direct Computer Agent chat |

Do not force exploratory or frequently changing work into one helper. Do not use free-form delegation for a sensitive operation that should have a narrow, testable contract.

## Generate a helper from a browser workflow

Use the Browser workflow recorder when a process must be learned from a website:

1. Make sure the Computer Agent's browser is signed in to the intended account.
2. Record one complete, representative workflow and explain the intended business outcome.
3. Stop the recording and start helper generation.
4. Use a neutral skill name when the website address contains a customer name or another detail that should not become the reusable skill identity.
5. Review the generation conversation and test every generated operation with safe data.

A recording is evidence, not proof that every variation works. Confirm the helper's inputs, result, failure behavior, and effect in the destination system.

Generation freezes the finalized recording IDs selected when it starts. If you capture better
evidence afterward, start a new generation. Follow the generation conversation and activity status
rather than treating a long-running phase as a failure.

## Build a Browser-independent helper

An API developer or an authorized setup Agent can add a focused helper to the skill that owns its instructions. Keep the implementation and usage guidance together so the capability can be installed, updated, and removed as one unit.

Declare only the managed credentials the helper needs. The helper receives those values at execution time; callers supply business inputs, never secret values.

The complete API walkthrough—including the helper source, managed-secret setup, registration,
execution, and conversational test—is in
[Workspaces, helpers, and browser automation](/docs/api-reference/agent-guides/computer-agent-workspaces-and-helpers#example-call-an-api-without-opening-the-browser).

## Test before exposing it to conversations

Confirm that the helper appears in the selected Computer Agent's helper list. Check its exact name, inputs, Browser requirement, owning skill, and required credential IDs. Run a safe representative request and inspect both the structured result and the destination system.

The runtime result and the registered result description must agree, including filters, durable IDs,
and normalized business fields callers rely on. Preserve opaque IDs exactly as returned.

For a conversational Agent:

* **Computer** can invoke an exact registered helper or delegate broader multi-step work.
* **Browser** is for ad hoc work in the live Browser, not registered helpers.

Add the Computer function and write instructions that map conversation values to the helper's exact
input names. A Computer Agent uses its own Computer automatically; select another Computer Agent only
when the work must be delegated across Agents. Test the staged conversational Agent before
publishing it.

Re-test Browser-backed helpers after the website changes. Re-test every helper after changing its owning skill or credentials.

<Note>
  Creating and operating helpers through the API? See [Workspaces, helpers, and browser automation](/docs/api-reference/agent-guides/computer-agent-workspaces-and-helpers).
</Note>