AI Provider for exo
This plugin registers exo as an AI provider in WordPress 7's AI Client SDK and Connectors page.
exo connects all your devices into an AI cluster, enabling you to run frontier models locally. It exposes an OpenAI-compatible API that this plugin connects to.
Features:
- Registers exo as a WordPress AI provider
- OpenAI-compatible text generation via exo's chat completions API
- Auto-detect active models from your running exo cluster
- Capability detection — displays model capabilities (Text, Code, Thinking, Vision) as badges
- "Connect & Detect" / "Save & Re-detect" connector flow
- Optional API key authentication with secure storage
- Configurable endpoint URL (default:
http://localhost:52415) - Settings integrated into WordPress 7's Connectors page
Choosing a Model:
exo exposes every model in its catalog, but only models actively loaded on your cluster will respond. Use "Connect & Detect" on the Connectors page to discover which models are running.
Recommended — Instruct models produce clean, usable output for WordPress AI features (title generation, content suggestions, etc.):
Llama-3.2-3B-Instruct-8bit— ~3 GB, fast, great for short tasksMeta-Llama-3.1-8B-Instruct-4bit— ~4 GB, good balance of speed and qualityLlama-3.3-70B-Instruct-4bit— ~35 GB, best quality, needs a larger cluster
Avoid — Reasoning/thinking models (Qwen3.5, DeepSeek, GLM, Nemotron-Nano) spend most tokens on internal chain-of-thought, producing slow responses with minimal visible output.
To load a model: exo run mlx-community/Llama-3.2-3B-Instruct-8bit
Requirements:
- WordPress 7.0 or later
- PHP 8.3 or later
- A running exo cluster (see exo documentation)
Configuration
The plugin can be configured via the Connectors page or environment variables:
AIPRFOEX_ENDPOINT— exo API endpoint (default:http://localhost:52415)AIPRFOEX_API_KEY— Optional API key for authenticationAIPRFOEX_MODEL— Model name to use (auto-detected if empty)
You can also define these as constants in wp-config.php.