MW Local AI Connector

MW Local AI Connector is a WordPress AI provider plugin plus a small local proxy for running inference against models on your own machine.

It is built for WordPress 7.0+ and the new Connectors primitives. The WordPress plugin runs on the remote site. The local proxy runs on a machine you control, talks to one or more OpenAI-compatible local backends such as Ollama or LM Studio, and can expose them to WordPress locally, through Tailscale Funnel, or through Cloudflare Tunnel.

This is an independent project by Matt Wiebe. It is not affiliated with, endorsed by, or sponsored by Actual Computer, Tailscale, Ollama, LM Studio, or any other third-party service it can be configured to talk to.

What It Does

• Registers a mwlai AI provider with the WordPress AI Client.
• Adds a Settings screen for entering the proxy URL, optional API key, and selected model.
• Loads model choices from the real proxied /v1/models endpoint.
• Restricts the provider to the selected model so WordPress auto-discovery consistently uses it.
• Ships a local node proxy that can:
  • scan common localhost ports for OpenAI-compatible providers,
  • persist provider configuration in ~/.config/mwlai-connector/.env,
  • expose the proxy through Tailscale Funnel or Cloudflare Tunnel,
  • protect tunneled requests with a bearer token.

Repository Layout

• mw-local-ai-connector.php: WordPress plugin bootstrap, settings UI, connector registration.
• src/: provider, model, and metadata directory classes.
• local/server.mjs: local multi-provider proxy and tunnel entrypoint.

Requirements

• WordPress 7.0+
• PHP 7.4+
• Node.js 20+
• One or more local OpenAI-compatible backends, such as:
  • Ollama
  • LM Studio
  • another localhost backend by port number
• Tailscale, if you want public exposure through Funnel
• cloudflared, if you want public exposure through Cloudflare Tunnel

WordPress Plugin Setup

1. Copy this plugin into wp-content/plugins/mwlai-connector.
2. Activate the plugin in WordPress.
3. Open Settings > Connectors or the dedicated Local AI settings page.
4. Enter:
   • the proxy endpoint URL,
   • the shared API key, when using a tunneled proxy,
   • the model to expose to WordPress.

The Connectors screen links back to this setup page.

Local Proxy Setup

Preferred install path:

npm install -g @mattwiebe/ai-connector-for-local-ai

Then initialize and run the proxy with:

laiproxy init
laiproxy up

macOS background service management:

laiproxy install
laiproxy start
laiproxy stop
laiproxy status
laiproxy rotate-key
laiproxy uninstall

You can also run it without a global install:

npx @mattwiebe/ai-connector-for-local-ai init
npx @mattwiebe/ai-connector-for-local-ai up

For local development from this repo, you can still use:

npm run init
npm run up
npm run service:install
npm run start
npm run stop
npm run status
npm run rotate-key
npm run service:uninstall

That guided setup will:

• scan common localhost ports for OpenAI-compatible providers,
• ask for any provider slugs and ports that were not detected,
• generate or accept an API key for tunneled modes,
• ask whether to run local-only, Tailscale Funnel, or Cloudflare Tunnel,
• ask which public Funnel port to use when Tailscale is selected, defaulting to 8443,
• save everything into ~/.config/mwlai-connector/.env when run through the laiproxy CLI.

When Funnel starts, the proxy also checks public DNS for the generated ts.net hostname. If public DNS has not propagated yet, WordPress may temporarily report Could not resolve host; Tailscale says Funnel DNS propagation can take up to 10 minutes.

Configured providers are exposed as one OpenAI-compatible endpoint. The proxy prefixes model IDs with the provider slug in /v1/models, for example ollama/llama3.2 or vibeproxy/openai/gpt-5-5. Requests should send that prefixed model ID; the proxy strips the slug before forwarding to the matching local provider.

Provider slugs and ports must both be unique. If two providers are configured with the same localhost port, startup fails with a configuration error instead of guessing which provider should own that port.

The persisted .env uses this provider format:

PORT="13531"
TUNNEL_MODE="local"
PROVIDERS="ollama:11434,lmstudio:1234,vibeproxy:3434"
API_KEY=""

The running proxy watches its .env file and restarts itself when that file changes. Local-only mode does not require an API key. Tailscale Funnel and Cloudflare Tunnel modes generate and require one.

After that, normal startup is non-interactive:

laiproxy up

To reconfigure later:

laiproxy init

On macOS, laiproxy install or npm run service:install writes a LaunchAgent at ~/Library/LaunchAgents/com.mattwiebe.mwlai-connector.plist so the proxy can keep running in the background across logins.

To rotate the shared API key for a tunneled proxy in the persisted .env file and print the new key:

laiproxy rotate-key

This also works as:

laiproxy rotate-key
npm run rotate-key

On macOS, if the LaunchAgent is currently running, rotate-key will restart it automatically so the background process picks up the new key immediately.

Useful overrides:

laiproxy up --port 13531
laiproxy up --provider ollama:11434 --provider lmstudio:1234
laiproxy up --tunnel cloudflare
laiproxy up --tunnel tailscale
laiproxy up --funnel-port 10000
laiproxy up --backend http://localhost:11434
laiproxy up --api-key your-secret
laiproxy up --no-tunnel

Development

Developer docs, release notes, packaging details, and reference material now live in docs/developing.md.