> ## Documentation Index > Fetch the complete documentation index at: https://docs.dubot.ai/llms.txt > Use this file to discover all available pages before exploring further. # Browser SDK API > Current public methods and properties on window.Dubot The CDN script installs one browser global: `window.Dubot`. ## Properties and inspection | Member | Type | Description | | --------------- | ---------------- | ---------------------------------------------------------------------------------------- | | `version` | `string` | Exact SDK build version; local builds report `dev` | | `initProps` | object or `null` | Last accepted, serializable initialization snapshot; excludes JWT and callback functions | | `actionChannel` | object | Published action-channel state and live presence connection state | | `getState()` | object | Combined version, initialization, and action-channel snapshot | ## Initialization ### `init(options?)` Initializes identity, configuration, placements, Resource Center, Guidance resume, analytics, and the optional action channel. Returns `Promise`. Await it before programmatic launches. Missing client tokens, configuration-fetch failures, duplicate calls, and calls after `destroy()` are reported in the console and can resolve without a working surface. Verify the rendered result and Network responses; resolution does not prove identity acceptance or successful rendering. | Option | Type | Notes | | ------------------------- | ---------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------- | | `userId` | `string` | Required for identified sessions and user-attributed analytics; omitted or empty means anonymous | | `email` | `string` | Optional identity attribution | | `identityToken` | `string` | Required when signed identity is enforced and for the end-user action channel; server-minted HS256 JWT with `sub = userId` | | `mode` | `bubble`, `inline`, `sidebar`, or `fullscreen` | Default for programmatic wizard rendering | | `position` | `bottom-left` or `bottom-right` | Call-site widget-position override | | `actions` | array | Initial complete client-side callback set | | `resourceCenterCallbacks` | array | Initial complete static-card callback set | | `context` | object | Flat scalar session context, approximately 2 KB maximum | The client token is read from the SDK script URL and is not an `init()` option. All other initialization options are optional. Omitting `mode` defaults programmatic rendering to `bubble`; configured placements keep their own presentation. An identical repeated call does not remount placements. A new token with otherwise unchanged options refreshes identity in place. Other changed options can restart initialization. Pass the complete intended configuration when reinitializing; `init()` is not a partial-update method. See [token renewal and account changes](/sdk/identity-context#renew-an-identity-token). ## Wizards and Resource Center The two presentation fields use different public values: | Field | Values | Purpose | | ----------------------------------- | ------------------------------------------------- | ----------------------------------------------- | | `init({ mode })` | `bubble`, `inline`, `sidebar`, `fullscreen` | Default for later programmatic Wizard rendering | | `renderAgent({ presentationMode })` | `Bubble`, `InlineWizard`, `Sidebar`, `Fullscreen` | Presentation for this rendered Wizard | | Method | Returns | Behavior | | ----------------------------------------------------- | ------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | `renderAgent({ slug, presentationMode?, selector? })` | `void` | Starts rendering a Wizard; `slug` is required and inline mode needs a matching `selector`. Fetching and mounting continue asynchronously; errors appear in the console or surface | | `openResourceCenter()` | `void` | Opens the mounted Resource Center when its trigger state allows it; has no effect when no Resource Center is mounted | | `closeResourceCenter()` | `void` | Closes the mounted Resource Center; has no effect when none is mounted | These methods use the current environment's configuration: the saved draft in a browser-extension Test session, or the published version in Production. They do not publish configuration. ## Host state | Method | Returns | Description | | --------------------- | ------- | ----------------------------------------------------------------------------------------------------------------------------------- | | `setActions(actions)` | `void` | Replaces the complete client-side callback set; pass `[]` to remove all callbacks | | `setContext(context)` | `void` | Replaces the complete flat scalar context map; pass `{}` to clear it. Nested values and overflow are dropped with a console warning | Each action has `tool_id: string`, `input_schema: object`, and `execute(input): unknown | Promise`. Each static Resource Center callback has `id: string` and `execute(): unknown | Promise`. See [SDK actions](/sdk/actions) for complete examples and the separate governance requirements. ## Guidance | Method | Description | | ------------------------------ | --------------------------------- | | `triggerGuidance({ guideId })` | Starts a published reviewed guide | | `triggerGuidance({ goal })` | Starts a goal-based run | The returned session has `stop()`, optional `guidanceSessionId`, and a `done` promise that resolves to an outcome containing `{ status, detail }`, where `status` is `completed`, `stopped`, or `error` and `detail` is a string. `done` resolves on failure as well; inspect its status instead of relying on a rejection. `stop()` returns `void`. Goal-based runs require workspace enablement and a controlled evaluation; see [Guidance availability and guide compatibility](/product/guidance). ## End-user MCP action channel | Method | Description | | -------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------- | | `listActionChannelGrants()` | Returns `Promise`; non-success HTTP responses return `[]`, while network or JSON parsing failures can reject | | `openActionChannelConnections()` | Returns `void`; opens the in-product grant and revocation surface | These methods return no grants or no-op when the SDK does not have both `userId` and `identityToken`. They belong to the [controlled-rollout end-user action channel](/integrations/mcp#end-user-action-channel). Each grant contains string fields `id`, `clientId`, `clientName`, `workspaceId`, `sub`, and `createdAt`. An empty list alone does not distinguish no grants from a rejected HTTP request; inspect the Network response when troubleshooting. ## Lifecycle ### `destroy()` Tears down the installed instance. It is safe to call more than once and terminal for that instance. Returns `void`. Reload the full page before initializing a fresh instance; reinserting the script into the same document does not revive it. `setAuthoringMode()` exists for the Dubot browser extension's authoring isolation. It is not a host-application integration API. For runtime diagnosis, see [Troubleshoot the browser SDK](/sdk/troubleshooting).