LLM Settings
LLM Settings is used to configure the site's AI services, including the service provider, keys, web tools, and generation parameters. Once configured, AI Chat in PageBuilder can generate content and answer questions properly.
Basic Settings
| Field | Description |
|---|---|
| Enable AI Features | The master switch for AI features. When disabled, the AI entry point no longer appears in the bottom-right corner of the editing interface |
After modifying the configuration, click Update to save; click Cancel to discard this change.
AI Provider Settings
Select an AI service provider and fill in the corresponding connection information.
| Field | Description |
|---|---|
| Provider | Select an AI service provider, see the table below |
| API Key | The API key for the selected provider. The input is masked; leave it empty if you're not changing it, and the saved key will be retained |
| API Endpoint | The service endpoint URL (the interface example shows https://api.openai.com). If using a self-hosted or proxy service, enter the corresponding address |
| Default Model | The default model name to call, such as gpt-4o, GLM-5.2, etc. |
Supported Providers
| Provider | Description |
|---|---|
| OpenAI | The official OpenAI service (GPT series) |
| Azure OpenAI | The OpenAI service hosted on Microsoft Azure |
| DeepSeek | DeepSeek models |
| Anthropic Claude | Anthropic's Claude series models |
| Google Gemini | The Google Gemini series models |
| AWS Bedrock | Multiple models hosted on Amazon Bedrock |
| Ollama | A locally deployed Ollama service, suitable for offline / private scenarios |
Relationship between Endpoint and Model
Different providers have different Endpoint and model name formats. For example, OpenAI defaults to https://api.openai.com with the model gpt-4o; if using a third-party service compatible with the OpenAI protocol, enter the corresponding address in Endpoint and the model name it supports in Model.
Web Tools Settings
Enable web access for the AI so it can fetch web pages or retrieve internet information, allowing it to answer more timely and broader questions.
| Field | Description |
|---|---|
| Web Tools | The master switch for web capabilities. When checked, it enables web_fetch (fetch web pages) and web_search (online search) |
| Search API Key | The Serper.dev API key required by web_search. The input is masked; leave it empty if you're not changing it, and the saved key will be retained |
The difference between the two tools
- web_fetch: Fetches the content of a specified URL for the AI to read (suitable for "summarize this web page"). No configuration required; works out of the box.
- web_search: Retrieves information on the internet by keyword; results come from Google (suitable for "find the latest information"). Requires the Search API Key above to use.
About Serper.dev
web_search calls the Google search interface through Serper.dev. Visit serper.dev to register an account and obtain an API Key; the free tier includes 2,500 queries per month, which is enough for everyday use.
Generation Parameters
Adjust the style and length of the AI output.
| Field | Description |
|---|---|
| Temperature | Controls randomness. 0 = focused (stable, conservative), 2 = creative (diverse, divergent). Default 0.7 |
| Max Tokens | The maximum number of tokens in the response. 0 = use the model's default maximum |
| Thinking Mode | Enables the model's extended thinking / reasoning capability. Applies to reasoning-capable models such as DeepSeek and Qwen3 — it improves answer quality but takes longer and consumes more tokens |
::: tips on choosing parameters
- Generating page content: A moderate Temperature (0.4–0.7) balances stability and creativity.
- Factual Q&A / SEO descriptions: A lower Temperature (0.2–0.4) gives more stable and controllable results.
- Long-form content: Increase Max Tokens appropriately to avoid the response being cut off.
- Complex reasoning questions: Enable Thinking Mode; for normal conversation, you can disable it to speed up the response. :::