Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/alibaba/page-agent/llms.txt

Use this file to discover all available pages before exploring further.

LLMConfig defines every setting that controls how Page Agent communicates with the LLM backend. Page Agent works with any OpenAI-compatible API that supports tool calls — including OpenAI, Anthropic-compatible proxies, Alibaba Cloud Qwen, and local models served by Ollama or LM Studio. These fields are all inherited directly by AgentConfig, so you never need to pass a nested object.
Never embed real API keys in client-side code. Browser JavaScript is publicly readable. Route LLM requests through a backend proxy and use customFetch to attach credentials server-side.

​
Fields

​
baseURL
string
required
The base URL of the OpenAI-compatible API endpoint. Do not include a trailing /chat/completions path — Page Agent appends the correct path automatically.
ProviderExample value
OpenAIhttps://api.openai.com/v1
Alibaba Cloud Qwenhttps://dashscope.aliyuncs.com/compatible-mode/v1
Ollama (local)http://localhost:11434/v1
LM Studio (local)http://localhost:1234/v1
Custom proxy/api/llm-proxy
​
model
string
required
The model identifier exactly as the provider expects it. For example gpt-4.1-mini, qwen3.5-plus, claude-sonnet-5, or qwen3:14b (Ollama tag format).
​
apiKey
string
API key for the LLM provider. Omit for local runtimes (Ollama, LM Studio) that do not require authentication, or when authentication is handled via customFetch.
​
temperature
number
deprecated
Sampling temperature. Deprecated — many models reject this parameter outright. Use transformRequestBody to set it only for models you have verified accept it.
​
maxRetries
number
default:"3"
Number of times the client will retry a failed LLM call before propagating the error. Applies to transient network errors and non-2xx responses.
​
disableNamedToolChoice
boolean
When true, the tool_choice field is removed from every request body. Required for LM Studio and some other local models that reject named tool choice with an error similar to "Invalid tool_choice type: 'object'".
​
transformRequestBody
(body: Record<string, unknown>) => Record<string, unknown> | undefined
Intercept and transform the complete request body immediately before it is sent to the provider. Return a modified copy of body, or mutate the argument in place and return undefined. Use this for provider-specific flags that are not part of the standard OpenAI schema.
// Add Claude-style prompt-caching hints
transformRequestBody: (body) => ({
  ...body,
  cache_control: { type: 'ephemeral' },
})
​
customFetch
typeof globalThis.fetch
Replacement fetch implementation used for all LLM API requests. The response must follow the OpenAI API response format. Common use cases:
  • Attach session cookies or bearer tokens without exposing them in source code
  • Route requests through a backend proxy to keep the API key server-side
  • Override CORS headers in a browser extension context
customFetch: (url, init) =>
  fetch(url, { ...init, credentials: 'include' })

​
Provider Examples

​
OpenAI

import { PageAgent } from 'page-agent'

const agent = new PageAgent({
  baseURL: 'https://api.openai.com/v1',
  model: 'gpt-4.1-mini',
  apiKey: 'sk-...',
})

​
Alibaba Cloud Qwen

import { PageAgent } from 'page-agent'

const agent = new PageAgent({
  baseURL: 'https://dashscope.aliyuncs.com/compatible-mode/v1',
  model: 'qwen3.5-plus',
  apiKey: 'sk-...',
})

​
Local Ollama

No API key is needed. The model name must match an Ollama tag you have already pulled. 
import { PageAgent } from 'page-agent'

const agent = new PageAgent({
  baseURL: 'http://localhost:11434/v1',
  model: 'qwen3:14b',
})
If you see "Invalid tool_choice type: 'object'" with a local model, add disableNamedToolChoice: true to your config.
Route all LLM traffic through your own server so the API key is never exposed to the browser. 
import { PageAgent } from 'page-agent'

const agent = new PageAgent({
  baseURL: '/api/llm-proxy',
  model: 'gpt-4.1',
  customFetch: (url, init) =>
    fetch(url, { ...init, credentials: 'include' }),
})

​
Claude Prompt Caching via Proxy

Use transformRequestBody to attach caching hints supported by your proxy or provider. 
import { PageAgent } from 'page-agent'

const agent = new PageAgent({
  baseURL: 'https://your-claude-proxy/v1',
  model: 'claude-sonnet-5',
  transformRequestBody: (body) => ({
    ...body,
    cache_control: { type: 'ephemeral' },
  }),
})

​
Full TypeScript Interface

export interface LLMConfig {
  /** Base URL of the OpenAI-compatible API (required) */
  baseURL: string

  /** Model identifier as accepted by the provider (required) */
  model: string

  /** API key; omit for local runtimes or when using customFetch */
  apiKey?: string

  /**
   * @deprecated No longer a standard parameter; many models reject it outright.
   * Use `transformRequestBody` to set it only for models you've verified.
   */
  temperature?: number

  /** Retry count on LLM call failure. Default: 3 */
  maxRetries?: number

  /**
   * Remove tool_choice from the request body.
   * Fix for LM Studio and some local models.
   */
  disableNamedToolChoice?: boolean

  /**
   * Transform the final request body before sending.
   * Return a new object or mutate and return undefined.
   */
  transformRequestBody?: (
    requestBody: Record<string, unknown>
  ) => Record<string, unknown> | undefined

  /**
   * Custom fetch for auth, proxying, or CORS overrides.
   * Response must follow OpenAI API format.
   */
  customFetch?: typeof globalThis.fetch
}

Build docs developers (and LLMs) love

Get started for freeTalk to us