SDK Reference

`BundleLLM.init(config?)`

Creates a new BundleLLM instance.

const ai = BundleLLM.init({ siteId: 'my-site' })

Parameter	Type	Description
`config.apiUrl`	`string` (optional)	OAuth redirect handler URL
`config.siteId`	`string` (optional)	Site identifier for analytics (auto-derived from domain if omitted)
`config.sessionTTL`	`number` (optional)	Session TTL in seconds. Stored credentials expire after this duration. No default (credentials persist indefinitely). Negative values are ignored.

Returns: BundleLLMInstance

Instance Methods

`renderSignIn(selector)`

Renders a provider picker in the target element. Shows OpenRouter (OAuth) and Anthropic (API key) with model selection.

ai.renderSignIn('#sign-in-container')

`renderChat(selector, options?)`

Renders a complete chat widget with provider picker, model selector, streaming, token usage, and sign-out.

ai.renderChat('#chat', {
  placeholder: 'Ask anything...',
  context: 'Page content here...',
  welcomeMessage: 'Connect your AI to start.',
  theme: 'light',
  markdown: true,
})

Option	Type	Default	Description
`placeholder`	`string`	`"Ask something..."`	Input placeholder
`context`	`string`		System prompt sent with every message
`welcomeMessage`	`string`	`"Connect your AI to start chatting."`	Shown before sign-in
`theme`	`"light" \| "dark"`	`"light"`	Color theme
`markdown`	`boolean`	`true`	Render markdown in responses. Set `false` for plain text.

See Chat Widget for full options.

`chat(request)`

Starts a streaming chat. Returns a ChatStream.

const stream = ai.chat({
  messages: [{ role: 'user', content: 'Hello' }],
  context: 'Optional system prompt...',
  model: 'anthropic/claude-sonnet-4',
  maxTokens: 4096,
  temperature: 0.7,
})

Parameter	Type	Description
`messages`	`Array`	Conversation history with `role` and `content`
`context`	`string` (optional)	Sent as system prompt with every message
`model`	`string` (optional)	Model override; uses selected model if omitted
`maxTokens`	`number` (optional)	Maximum response tokens (default: 4096)
`temperature`	`number` (optional)	Sampling temperature

Returns: ChatStream

`setContext(context?)`

Updates the system prompt context dynamically. For renderChat, this overrides the initial context option. For direct chat() calls, this is used as a fallback when request.context is not provided. Pass undefined to clear the override and revert to the original context.

// Update context (e.g., when page content changes)
ai.setContext('Updated page content...')

// Clear the override
ai.setContext(undefined)

Parameter	Type	Description
`context`	`string \| undefined`	New system prompt context, or `undefined` to clear

Context exceeding 10,000 characters is truncated with a console warning.

`getStatus()`

Returns the current connection status.

const status = ai.getStatus()
// { connected: true, provider: 'anthropic', model: 'claude-sonnet-4-20250514' }

Returns: { connected: boolean, provider?: string, model?: string }

`setModel(modelId)`

Changes the active model while connected. Persists to localStorage.

ai.setModel('anthropic/claude-haiku-4')

`getModels()`

Returns available models for the connected provider.

const models = ai.getModels()
// [{ id: 'claude-sonnet-4-20250514', name: 'Claude Sonnet 4' }, ...]

Returns: Array<{ id: string, name: string }>

`disconnect()`

Signs out the user’s AI provider. Clears stored connection.

ai.disconnect()

`on(event, callback)`

Listens for events.

ai.on('connected', ({ provider, model }) => { ... })
ai.on('disconnected', () => { ... })
ai.on('error', ({ message }) => { ... })

Event	Callback Data	Description
`connected`	`{ provider, model }`	User connected a provider
`disconnected`	.	User signed out
`error`	`{ message }`	Error occurred

`off(event, callback)`

Removes an event listener.

`destroy()`

Tears down the instance completely. Removes rendered widget and sign-in DOM elements, clears the connection, removes event listeners, and cleans up internal state. All subsequent method calls on the instance will no-op. chat() will emit an "Instance has been destroyed" error.

ai.destroy()

ChatStream

Returned by ai.chat(). Supports chained .on() calls.

stream
  .on('start', ({ messageId }) => { ... })
  .on('delta', (text) => { ... })
  .on('done', ({ usage }) => { ... })
  .on('error', ({ message }) => { ... })

Event	Callback Data	Description
`start`	`{ messageId }`	Stream started
`delta`	`string`	Text chunk received
`done`	`{ usage? }`	Stream complete; `usage` has `inputTokens` and `outputTokens`
`error`	`{ message }`	Stream error

`stream.cancel()`

Cancels the active stream.

stream.cancel()

Utilities

`BundleLLM.renderMarkdown(text)`

Converts markdown text to styled HTML. Available for custom UI integrations where you want the same formatting as the drop-in widget.

stream.on('delta', (text) => {
  fullText += text
  messagesEl.innerHTML = BundleLLM.renderMarkdown(fullText)
})

Supports: code blocks (with language labels), inline code, bold, italic, headers, ordered/unordered lists, and links. HTML in the input is escaped. Links are restricted to http: and https: protocols.

Returns: string (HTML)