LLM Providers

LLM Providers

OpenWP supports 4 LLM providers through a unified interface. Provider clients are created via the Provider_Factory and implement ProviderClientInterface.

Provider Factory (inc/Providers/Provider_Factory.php)

$client = Provider_Factory::create('openai');    // OpenAI_Client
$client = Provider_Factory::create('anthropic'); // Anthropic_Client
$client = Provider_Factory::create('glm');       // GLM_Client
$client = Provider_Factory::create('openrouter');// OpenRouter_Client

Provider Interface

All providers implement ProviderClientInterface with two methods:

interface ProviderClientInterface {
    public function generate(ProviderRequest $request): ProviderResponse|WP_Error;
    public function generate_stream(ProviderRequest $request, callable $on_delta): ProviderResponse|WP_Error;
}

Providers

OpenAI (inc/Providers/OpenAI_Client.php)

• API: Responses API
• Default Model: gpt-5.2
• Streaming: Supported via Stream_Transport

Anthropic (inc/Providers/Anthropic_Client.php)

• API: Messages API
• Default Model: claude-3-5-sonnet-latest
• Streaming: Supported via Stream_Transport

GLM (inc/Providers/GLM_Client.php)

• API: Z.AI Chat Completions
• Default Model: glm-5
• Streaming: Supported via Stream_Transport

OpenRouter (inc/Providers/OpenRouter_Client.php)

• API: Unified gateway (compatible with OpenAI format)
• Default Model: anthropic/claude-sonnet-4-5
• Streaming: Supported via Stream_Transport

Request / Response Objects

ProviderRequest (inc/Providers/ProviderRequest.php)

$request = new ProviderRequest([
    'system_prompt' => '...',    // System instructions
    'prompt'        => '...',    // User prompt
    'model'         => 'gpt-5.2',
    'schema'        => [...],    // Expected JSON output schema
    'stream'        => true,     // Enable streaming
    'temperature'   => 0.2,      // Low for predictable JSON output
    'timeout'       => 120,      // Seconds
    'max_tokens'    => 16384,    // Max output tokens
]);

ProviderResponse (inc/Providers/ProviderResponse.php)

$response->output_text;    // Raw text output from LLM
$response->input_tokens;   // Input token count
$response->output_tokens;  // Output token count

Stream Transport (inc/Providers/Stream_Transport.php)

Handles HTTP streaming for all providers:

• Uses wp_remote_request with streaming enabled
• Parses SSE data: lines in real-time
• Forwards text deltas to callback
• Handles connection timeouts

API Key Storage

Provider API keys are:

1. Encrypted via Encryption_Service (Sodium or AES-256-CBC)
2. Stored in OPENWP_OPTION_PROVIDER_KEYS option
3. Managed by Provider_Key_Manager
4. Masked for UI display (only last 4 characters shown)

Configuration

Default provider and model are configurable per-provider:

Setting	Key	Default
Default provider	default_provider	openrouter
OpenAI model	default_model_openai	gpt-5.2
Anthropic model	default_model_anthropic	claude-3-5-sonnet-latest
GLM model	default_model_glm	glm-5
OpenRouter model	default_model_openrouter	anthropic/claude-sonnet-4-5
Request timeout	timeout_seconds	45
Temperature	(hardcoded)	0.2
Max tokens	(hardcoded)	16384