Skip to main content

Run AI locally (Ollama, LM Studio)

Auto RepAI can talk to a model running on your own machine instead of a cloud provider. There is no API key to buy, no per-word cost, and no prompt or article ever leaves your server.

This is in the free plugin — nothing here needs Pro.

When this is the right choice

Building a site locally, working with client material you'd rather not send to a third party, generating a lot of draft content, or simply not wanting a metered bill while you experiment.

What you need​

Any server that speaks the OpenAI chat API. The common ones:

ToolDefault addressNotes
Ollamahttp://localhost:11434Easiest starting point on Mac, Windows and Linux
LM Studiohttp://localhost:1234Desktop app with a model browser; enable the local server
vLLMhttp://localhost:8000For serious hardware and concurrent requests
LiteLLMvariesA gateway in front of many providers at once

Step 1 — Install a model​

With Ollama, one command pulls a model:

ollama pull llama3.2

Small models (3B–8B) are fast on ordinary laptops and perfectly capable of outlines, product copy, FAQ sections and rewrites. Larger models (14B and up) write noticeably better long-form articles but need more memory.

Check it is running:

curl http://localhost:11434/v1/models

Step 2 — Point Auto RepAI at it​

  1. In WordPress, go to Auto RepAI → Settings → API Keys.
  2. Find the Local AI server section.
  3. Server URL — http://localhost:11434 (the /v1 part is added for you).
  4. Model name — exactly as your server names it, e.g. llama3.2:latest.
  5. API key — leave empty. Local servers don't use one.
  6. Click Test local server.

A successful test reports the models the server actually has installed, for example: Connected. 3 model(s) available: llama3.2:latest, qwen2.5:14b, mistral-nemo:12b.

Then click Save Settings.

Step 3 — Use it​

Open Auto RepAI → Content Generator and pick Local AI server in the provider list. The model dropdown fills itself from your server, so pulling a new model in Ollama makes it selectable here without touching the plugin.

Generate as usual. The entry in Auto RepAI → History will show a cost of 0 — because there was nothing to bill.

Building a whole site offline​

This pairs well with local development. A typical agency workflow:

  1. Spin the site up locally (for example in Studio, the free local WordPress app).
  2. Run Ollama on the same machine.
  3. Use Bulk Content to turn a topic list into drafts, and Auto Repair to tighten what's already there — all offline, at no cost, with the client's material never leaving your laptop.
  4. Switch the provider to OpenAI later if you want a stronger model for the pieces that matter most.
Moving a local site to production

Images generated locally are stored with a local URL such as http://localhost:8881/wp-content/uploads/.... That is normal for any local WordPress site — remember to run a proper search-replace for the site URL when you migrate, or the images will break. Migration tools do this for you.

What local models can and can't do here​

WorksDoesn't
Content Generator, Bulk Content, Auto Repair, GEO rewrites, product copy — anything textImage generation — use OpenAI for that
Any model your server exposesAudio transcription and speech

Text is where local models shine. Image models are a different kind of server and Auto RepAI keeps image generation on the providers that support it.

Troubleshooting​

"No local model name is configured yet." The model field is empty, or the name doesn't match. Run ollama list and copy the name exactly, tag included.

"The server did not return a model list." The server isn't reachable at that address. Check it is running, and that the port matches.

Connection works but generation is slow. Small models are far quicker; a 3B model on a laptop answers in seconds, a 30B model may take minutes. Model size, not the plugin, is the bottleneck.

Running Ollama on another machine. Use its LAN address (for example http://192.168.1.20:11434) and make sure Ollama listens on the network (OLLAMA_HOST=0.0.0.0).

See also​