Set up a new workspace
A workspace that has not been configured yet is set up from the chat, in two stages, with the browser extension as the assistant's way of reading your site.
A new workspace requires onboarding before it accepts configuration changes. Until onboarding is complete, write operations return an error and reads stay available. setup:status reports where the workspace is and what comes next; the assistant calls it first.
Onboarding has two stages.
1. The spec
The assistant works through a set of questions about your company, your website, the forms and flows the site already has, and what the agent is for. It asks them in order, a part at a time. Some parts need evidence from the site itself rather than your answers.
The browser extension
The assistant reads your site through Algomo's browser extension. As someone on your team browses the site with the extension enabled on it, the extension captures the pages and the data calls those pages make. That is what the assistant uses to see what the site actually does, and it is the only reliable way to see it on a modern site.
Install the extension, sign in with your Algomo account, and enable it on your website's domain. Spec parts that need site evidence accept answers only once a tab is enabled there. If you do not want to install it, say so; the assistant records that and those parts are answered from what you tell it instead.
2. The build plan
From the spec, the assistant proposes a build plan: the parts it will build, each grounded in something from the spec. You approve the plan in the chat. Submitting it completes onboarding and turns each part into a task.
The assistant then works through the tasks: website access verified against the real pages, a greeting, tools that read the site, cards where data is visual, and a final round of test conversations. It reports back with links to what it built, and you review and publish in the dashboard like any other session.
Recipes
Each capability (greetings, goals, website browsing through connectors, cards, booking calendars, search, human handoff, custom agents) has a recipe: a playbook with its best practices and, for most, a template that stages a known-good starting point into a session. The assistant reads the recipes before it proposes the plan; you can ask it for any of them.