---
title: "Browser Use in Workflows"
canonical: https://trickest.com/docs/using-the-app/workflow-and-executions/using-browser-nodes
description: "Check a page with Browser Use, inspect its evidence, and continue on the same computer."
---

# Browser Use in Workflows

Use a Browser Use node when a workflow needs to interact with a rendered web page and verify what it shows. Start with the three-node asset review example, or follow the single-page check below to learn the controls.

## Before you start

You need an Enterprise vault, permission to edit and execute a workflow, an execution fleet, and an available Browser Use tool. Choose a computer you can use, or a saved snapshot from which the run can create a separate computer. Your deployment administrator must configure the interactive runtime and a computer CLI that supports same-computer continuation.

A Library entry alone does not establish that the tool is runnable. If **Computer** lists no usable source, resolve that before starting. The small check below uses `https://example.com`; the workflow example uses the synthetic practice page.

## Example: review high-risk assets

Use three Browser Use nodes to check that a saved filter survives a page reload. The [practice page](/docs/examples/interactive-workflows/workspace.html) contains three synthetic assets and stores its settings in the browser. It needs no account or external API.

<Frame editorial maxWidth="780px" darkSrc="https://res.cloudinary.com/db14crach/image/upload/v1792764746/www/docs/connectors-workflows-2026-10-23/product/browser-asset-review-results-dark-a25de2e9d7b0692b.png" caption="A completed Browser Use workflow with saved page screenshots. All three nodes passed their checks.">
  <img src="https://res.cloudinary.com/db14crach/image/upload/v1792764744/www/docs/connectors-workflows-2026-10-23/product/browser-asset-review-results-light-cc8c7f20f0ced545.png" alt="Completed Browser Use workflow showing the original inventory, two filtered assets and the same filter after reload; each node says Checks passed." />
</Frame>

1. **Open inventory.** Open the practice page and click **Reset demo**. Check that `#asset-count` equals `3 assets`, then save a screenshot named `inventory-start`.
2. **Filter high risk.** Continue the current page and click **High risk**. Check that `#asset-count` equals `2 assets`, then save `high-risk-assets`.
3. **Verify after reload.** Continue the current page and reload it. Check that `#asset-count` still equals `2 assets` and `#density` equals `Comfortable`, then save `filter-persisted`.

<Frame editorial maxWidth="760px" caption="The final Browser Use node saved this screenshot after reloading. The two high-risk assets remain selected.">
  <img src="https://res.cloudinary.com/db14crach/image/upload/v1792764751/www/docs/connectors-workflows-2026-10-23/product/browser-filter-persisted-result-2c227d5a3d380e76.png" alt="Actual Browser Use output showing two high-risk assets and the Comfortable view after reload." />
</Frame>

### Import and configure the example

1. Download the [asset review workflow](/docs/examples/interactive-workflows/browser-asset-review.workflow.json).
2. Open an empty workflow. Right-click the canvas, select **Import workflow…**, then **Upload a .json file**. Import adds nodes to the current canvas.
3. Open **Open inventory > Setup** and choose your computer or snapshot. Under **Steps > Write steps**, set **Page URL** to the address of the practice page above in your deployment, then select **Save steps**.
4. Check that **Filter high risk** continues after **Open inventory**, and **Verify after reload** continues after **Filter high risk**. The download includes these connections and each node's inline steps.
5. When ready, select **Execute** to run the whole workflow. Inspect each node's `result.jsonl` assertions and saved screenshots. If the last check fails, compare the filter before and after reload; a completed job alone does not establish persistence.

The download deliberately leaves the first computer unselected. The default page URL points to the public practice page. If you host the practice page elsewhere, update this URL before running. Separate recipe files are available for [opening the page](/docs/examples/interactive-workflows/open.recipe.json), [filtering](/docs/examples/interactive-workflows/filter.recipe.json), and [checking persistence](/docs/examples/interactive-workflows/verify-filter.recipe.json).

The graph and recipes pass the v1.2.1 import and configuration validators. This example completed on a dedicated demo computer: all three nodes passed, including the assertions in each `result.jsonl`. To include desktop interaction in the same sequence, see the [Computer Use example](/docs/using-the-app/workflow-and-executions/using-computer-nodes#example-change-a-desktop-view-and-verify-it).

## Add the node and choose its computer

1. Open a workflow, select **Add node**, and find **Browser Use**.
2. Open the node and show **Setup** if its side panel is collapsed.
3. Under **Computer**, select **Choose a computer** and select the source.
4. In **Start from**, choose **Current state** to use that computer as it is, or a dated saved snapshot to start a separate computer.

Using current state leaves the selected computer running after the run. A computer created from a saved snapshot stops at the end or on failure. A snapshot can remain selected even when its original computer is unavailable; choose an available source before changing it.

<Frame editorial maxWidth="360px" darkSrc="https://res.cloudinary.com/db14crach/image/upload/v1792713990/www/docs/release-1-2-1-2026-10-23/product/browser-setup-dark-5a0d52eb3976771d.png" caption="The saved example retains a snapshot whose source is unavailable. Check the source and snapshot before running.">
  <img src="https://res.cloudinary.com/db14crach/image/upload/v1792713992/www/docs/release-1-2-1-2026-10-23/product/browser-setup-light-fd1da28bfe49cdaa.png" alt="Browser Use Setup with Choose a computer, Continue after, Computer and Start from controls." />
</Frame>

## Write a small check

Under **Steps**, select **Write steps**. Set **Start** to **Open a page** and **Page URL** to `https://example.com`.

Use **Add step** and expand each step to configure it:

| Step action | Fields |
| --- | --- |
| **Check text** | **CSS selector:** `h1`; **Expected text:** `Example Domain` |
| **Save screenshot** | **Screenshot name:** `page-check` |

The action keys in recipe files are `assert_text` and `screenshot`. Use the action descriptions in your installed tool if the visible label differs. Select **Save steps** before running. Closing the panel keeps an unsaved draft; it does not save those steps for execution. Use **Discard changes** to return to the saved steps.

<Frame editorial maxWidth="760px" darkSrc="https://res.cloudinary.com/db14crach/image/upload/v1792713994/www/docs/release-1-2-1-2026-10-23/product/browser-step-controls-dark-88c0891dc470f2b3.png" caption="Choose inline steps or a recipe file. Save inline changes before starting a run.">
  <img src="https://res.cloudinary.com/db14crach/image/upload/v1792713996/www/docs/release-1-2-1-2026-10-23/product/browser-step-controls-light-6992a14edb5c1129.png" alt="Browser Use step source controls with Write steps, From a file, and Start set to Open a page." />
</Frame>

For an authenticated page, **Check the signed-in account first** lets you set an account selector and expected account text. Use it to catch an unexpected session before taking actions. It verifies the visible account indicator; it does not sign you in.

## Use a recipe file instead

Download the [browser check recipe](/docs/examples/release-1.2.1/browser-check.recipe.json), or create a JSON file containing:

```json
{
  "version": 1,
  "url": "https://example.com",
  "steps": [
    {"action": "assert_text", "selector": "h1", "equals": "Example Domain"},
    {"action": "screenshot", "name": "page-check"}
  ]
}
```

Select **From a file > Connect a recipe** and supply that one JSON file, either from uploaded storage or a script's file output. A results folder from another Browser Use node is not a recipe. Use inline steps or a file for a node, rather than connecting both.

## Run and inspect the evidence

Run the node and open its **Browser**, **Logs**, and **Files** views. A running top-level node can show the live read-only browser; a completed node can show a screenshot saved by that run. Missing preview evidence is not a successful check.

In the results folder, inspect `result.jsonl` and the screenshots. Check the assertion outcome as well as the node's execution status. A job can finish while reporting an assertion failure or a missing login. If the assertion fails, compare the screenshot with the expected text and selector before changing the check.

## Continue in another node

Add a second Browser Use or [Computer Use](/docs/using-the-app/workflow-and-executions/using-computer-nodes) node. In the second node's **Computer** section, choose **Continue after…** and select the first node. Set its own steps to **Continue current page** or **Continue current screen** when it should preserve the current view.

For a second browser check, use the [continuation recipe](/docs/examples/release-1.2.1/browser-continue.recipe.json): reload, verify the same heading, and save `after-reload`.

Run the **whole workflow** to test a continuation chain. It is sequential and cannot be distributed. One computer can continue to one next node, and the next node waits for the previous checks to pass. Testing only the later node cannot reuse a handoff already consumed by another execution.

## If the node cannot run

Check plan access, tool availability and the selected source. Missing runtime configuration requires your deployment administrator. An invalid recipe needs correction before execution; a failed assertion needs investigation in `result.jsonl`. A disconnected live viewer does not establish that the workflow stopped, so inspect run status before retrying.

The recipe files pass the released recipe validator. The example images show a completed platform workflow and its saved output. The control screenshots below the example show configuration.

---
_Markdown source of https://trickest.com/docs/using-the-app/workflow-and-executions/using-browser-nodes._
