# Build and improve your app

Ask for a change, wait for the update, and review the result.

## What you’ll have at the end

A focused change you can review in Preview and test against your request.

Source: https://docs.rumata.dev/guides/prompting
Reviewed: 2026-11-01

<a id="planning"></a>

## Follow the first build

Let Rumata finish the first version before you review it.

After you approve the plan, follow progress and completed work in the chat. The preview may appear while Rumata is still making changes. Rumata’s live AI agent also tests your application in its own Chrome browser, interacting with the app to check that features work as expected and identify issues. You can watch this testing live in the Agent browser view, like a broadcast from the agent’s Chrome window, and follow what it clicks, types, and checks.

Follow the live testing as the agent checks your app and works through any issues it finds. Wait for Rumata to say the application is ready to try, and answer any questions it asks along the way.

> **Check your result**
> Open the preview and compare it with your prompt. If there is no completion reply, check for an unanswered question or agent sign-in card; if the preview cannot connect, follow [troubleshooting](https://docs.rumata.dev/guides/troubleshooting#preview).

Learn more: [Chat progress and decisions](https://docs.rumata.dev/guides/controls-chat-decisions)

<a id="request-change"></a>

## Make changes and manage your conversations

Your first version is just the beginning! With Rumata, you can continuously improve your application by chatting with the AI agent.

Request new features, adjust the design, fix issues, or change how something works—all through natural language.

1. Open your application in Rumata.
2. In the chat, describe what you'd like to change. Be specific about the feature, expected result, and anything you want to keep unchanged.
3. Click Send to submit your request.
4. Follow the agent's progress and answer any questions or approval requests if needed.
5. Once the update is complete, use Preview Mode to test the changes.

> **Example: Example prompt**
> Add a search bar to the customers page so users can quickly find customers by name or email. Keep the existing filters and sorting options working.

Rumata supports multiple chat sessions within the same factory, allowing you to organize conversations without creating a new application.

Each session keeps its own conversation history, while all sessions work on the same underlying application.

You can use different sessions for different tasks, such as improving the design, adding new features, or fixing issues.

To work with sessions:

1. Open the chat sidebar in your factory.
2. Select an existing session to continue a previous conversation, or create a new session to start a fresh one.
3. Describe what you'd like to work on. Your changes will apply to the same application.

> **Good to know**
> Starting a new session doesn't create a new application. Sessions help you organize your conversations while keeping your work within the same factory.

Once Rumata confirms that your changes are complete, open Preview Mode and test the updated feature.

If something isn't quite right, describe what needs adjusting and continue refining your application.

Learn more: [Chat and session controls](https://docs.rumata.dev/guides/controls-chat)

<a id="review-change"></a>

## Review the update

Check whether the change solves the problem.

Read Rumata’s reply for a description of the changes, then open [Preview Mode](https://docs.rumata.dev/guides/controls-preview#navigation) to try the current development version.

1. Open the page or feature you asked to change. Use a test account if the task requires sign-in.
2. Try the changed behavior and compare the actual result with what you requested.
3. Repeat the application’s main task and check any saved information or access rules affected by the change.
4. Refresh Preview if it shows an older version. Report any unexpected behavior and retest after the fix.

> **Optional example**
> For a portfolio filter, choose a project type and check the displayed projects. Clear the filter and open a project to confirm the original links still work.

> **Check your result**
> Expected result: the requested change works and existing functionality still behaves as intended. Continue directly to [Test](https://docs.rumata.dev/guides/testing). [Understand your app](https://docs.rumata.dev/guides/understand-project) is optional.

Learn more: [Preview navigation and refresh](https://docs.rumata.dev/guides/controls-preview)

<a id="worked-example"></a>

## Optional worked example: Collab

This optional record illustrates the steps above using Collab Creator Management, an application for brands and creators. You do not need this factory, its roles, or its features to follow the manual. On September 29, 2026, its Factory bootstrap conversation contained the submitted requests, approval, and completed replies below. Historical test output and the September 29 browser checks are distinguished explicitly; this example has no verified live release.

| Stage | Evidence and next check |
| --- | --- |
| Submitted idea | The original submitted prompt requests Collab for brands and agencies, including four roles, creator CRM, campaigns, content review, finance tracking, and AI. A submitted answer chooses separate private workspaces for each brand or agency. |
| Plan approval | The plan is titled “Collab — first-version product foundation.” The next submitted message says “Implement the approved plan above now. Complete the work, test it, and verify the result.” This verifies explicit approval in chat, not a capture of active approval buttons. |
| Completed build | The completed reply reports connected pages, four roles, persistent data, campaign collaboration, and five passing tests. It also states that external provider integrations are unconnected and payments/signatures use manual tracking. The current preview renders Collab’s sign-in screen. |
| Improvement | A sent request asks for each admin to fund their own AI use. The completed reply reports caller-funded AI, protected Manager/Admin access, a $0.15 per-run cap, five creator recommendations, and six campaign insights. That cap belongs to this application; it is not a Rumata-wide limit. |
| Test | The recorded bun run test output shows one test file and five tests passed. In the current browser retest, Brand / Agency demo sign-in returned HTTP 500 with “Authentication failed.” The dashboard, customer task, and AI improvement could not be retested. Resolve sign-in, then repeat Test before publishing. |
| Publication | The first-deployment dialog showed Not Deployed, no release history, and 15 workflows to deploy with 0 already live. Review deployment listed the caller-funded AI warning. No deployment was started. Follow Publish after a successful retest, confirm the reviewed source selection and deployed revision, then verify the live task. |

Next: [Test the customer task](https://docs.rumata.dev/guides/testing#customer-task). If sign-in or another step fails, [report the failing step](https://docs.rumata.dev/guides/testing#fix). Continue to [Publish](https://docs.rumata.dev/guides/publishing#prepare) only after the retest succeeds.

![Completed Collab build reply reporting connected features, tests, and remaining integration limits.](https://docs.rumata.dev/media/manual/collab-build-complete.jpg)

The completed first-build reply in Collab’s history. It reports passing checks and explicitly leaves external integrations unconnected; this is not evidence of a published website.

![Sent request asking for each admin to pay for their own AI use, followed by its Sent marker.](https://docs.rumata.dev/media/manual/collab-submitted-improvement.jpg)

Optional example: Collab’s submitted improvement request asks to change who pays for AI use. Its roles and funding requirements belong to this application.

![Completed caller-funded AI reply reporting protected access, the application’s spending cap, recommendations, and tests.](https://docs.rumata.dev/media/manual/collab-improvement-complete.jpg)

The recorded completion reply reports caller-funded AI and successful checks. Its $0.15 cap is specific to Collab. The current demo sign-in failure prevents a fresh end-to-end verification.

![Actual saved test output: one test file and five tests passed for collab.test.ts.](https://docs.rumata.dev/media/manual/collab-completed-tests.jpg)

Completed bun run test output from the improvement’s recorded tool call: 1 file and 5 tests passed. These are historical code checks, not a passing result for today’s browser retest.

![Collab sign-in form with empty credentials, demo role buttons, and Unable to open demo. Please try again.](https://docs.rumata.dev/media/manual/collab-demo-sign-in-error.jpg)

Actual September 29 retest: choosing Brand / Agency failed. The Network panel recorded HTTP 500 with “Authentication failed.” Fix and retest before treating the customer journey as complete.

## Related guides

- [Chat, sessions, models, and file controls](https://docs.rumata.dev/guides/controls-chat)
- [Preview, network, and console controls](https://docs.rumata.dev/guides/controls-preview)