页面 · Laravilt

Providers

Configure the AI providers that power chat, agents and search.

Every provider implements Laravilt\AI\Contracts\AIProvider and extends Laravilt\AI\Providers\BaseProvider.

Provider	Class	Default model	Page
OpenAI	`OpenAIProvider`	`gpt-4o-mini`	OpenAI
Anthropic	`AnthropicProvider`	`claude-sonnet-4-20250514`	Anthropic
Google Gemini	`GeminiProvider`	`gemini-2.0-flash-exp`	Gemini
DeepSeek	`DeepSeekProvider`	`deepseek-chat`	DeepSeek
Perplexity	`PerplexityProvider`	`sonar`	Perplexity

Config file

config/laravilt-ai.php (published by php artisan laravilt:ai:install):

php

Each provider reads api_key, model, base_url, temperature and max_tokens from its entry.

Using the AI manager

Laravilt\AI\AIManager builds a provider for every configured entry that has an API key.

php

AIManager builds providers from config only for openai, anthropic, gemini and deepseek. To use Perplexity, register it on the panel with AIProviderBuilder or add it with $ai->addProvider(new PerplexityProvider).

Panel providers

A panel lists its providers with aiProviders(). The configure callback receives the provider instance:

php

Shortcuts: openai(), anthropic(), gemini(), deepseek(), perplexity().

Provider methods

Method	Description
`apiKey(string)`	Set the API key
`model(string\|BackedEnum)`	Set the model (string or model enum)
`baseUrl(string)`	Override the API base URL
`temperature(float)`	Sampling temperature
`maxTokens(int)`	Maximum output tokens
`enabled()` / `disabled()`	Toggle the provider
`chat($messages, $options)`	Returns `['content' => ..., 'usage' => [...]]`
`chatWithTools($messages, $tools, $options)`	Adds `tool_calls` to the result
`streamChat($messages, $options)`	Generator of text chunks
`streamChatRealtime($messages, $callback, $options)`	Calls `$callback` for each chunk as it arrives

Per-call $options accept model, temperature and max_tokens (OpenAI and DeepSeek also accept tool_choice in chatWithTools()).

Model enums live in Laravilt\AI\Enums: OpenAIModel, AnthropicModel, GeminiModel, DeepSeekModel, PerplexityModel.