Create a Web Agent
The Quick start wizard creates a Web Agent from your app's code and starts its first analysis. It asks for the framework, a name, the code, and how the agent follows new commits.
Before you begin
- Your workspace has a model provider. The analysis runs on it, and the wizard is locked with Connect your AI provider first until one is set up. See Connect your model provider.
- You can reach the app's front-end code in GitHub, GitLab or Bitbucket, or you have it as a ZIP file.
- The app is built with one of the supported frameworks.
Create the agent
To create a Web Agent:
- In the portal, go to Web Agents and select New Web Agent.
- Under Choose your app type, select the framework your app is built with. Single Page Application lists the client-rendered frameworks, and Server Side Rendering lists Next.js.
- Under Details, enter the Assistant name. The name identifies the agent in the portal and is the first title of the widget, which you can change later in Look & feel.
- Under Environments, select the environments the agent has. To add one of your own, enter its name and select Add. Then select Continue.
- Under Connect your code, select where the code is and choose the repository. See Connect your code or Upload a ZIP instead of a repository.
- Under Keep the assistant up to date, select Automatic or Manual, then Continue.
- Automatic updates the agent on every push to the connected branch. It is the recommended option because the actions follow your code without anyone starting a run.
- Manual leaves the updates to you.
- Under Extra resources, add reference repositories or API docs, or select Skip for now. See Add reference repositories and API docs.
- Under Review & create, check the summary and select Create & start analysis.
The portal opens the analysis page and the run starts. The analysis uses tokens from your model provider, and a larger codebase uses more.
Note
Staging is selected when the step opens. New versions are served on the first environment in your selection. The install snippet names no environment. It loads Production, and while the agent has no Production it loads the agent's first environment. To have your site run on Production from the start, clear Staging and select Production. See Versions, environments and deployments.
What to do while the analysis runs
The analysis stops once to ask which modules to analyze. See Choose what to analyze. You can leave the page while it runs and come back from the agent's row in Web Agents.
When the summary appears, select Continue to setup. The Set up your assistant wizard goes through the persona, the actions, web search, the knowledge base, the look and the chat model, and ends on the install snippet. Selecting Launch publishes the first version.
Troubleshooting
| Symptom | Cause | Fix |
|---|---|---|
| The page shows Connect your AI provider first | The workspace has no working model provider. | Set one up in Settings > Model Provider. |
| The page shows Installation requested after you return from GitHub | You cannot install the Faustine GitHub App on that organization, so GitHub asked its owners to approve it. | Wait for an owner to approve, then come back and select the connection. The wizard keeps what you entered. |
| Couldn't create the assistant appears under the summary | The message after the colon is the reason the service gave. | Fix what the message names and select Create & start analysis again. |
Next steps
- The analysis: what each stage does and what it produces.
- Choose what to analyze: pick the modules that become actions.
- Actions: review what the analysis generated.
- Installation: add the agent to your site.
Last updated November 11, 2026