AI providers

AI is optional. Without it, SimplifyGuide writes every step from its own rules (“In the left menu, go to Posts → Add New”), which read well on most WordPress screens. AI polishes that wording, rewrites single steps, and turns voice narration into notes.

Providers are set up under SimplifyGuide → Settings → AI writing, in the Providers list at the bottom of the tab.

The providers

Provider	Writes steps	Transcribes voice	Default model	Where to get a key
WordPress AI (site connection)	Yes	No	Chosen in WordPress	No key here; uses Settings → Connectors
Anthropic (Claude)	Yes	No	claude-opus-5	console.anthropic.com/settings/keys
OpenAI	Yes	Yes (whisper-1)	None — pick after checking the key	platform.openai.com/api-keys
DeepSeek	Yes	No	deepseek-chat	platform.deepseek.com/api_keys
OpenRouter	Yes	No	None — pick after checking the key	openrouter.ai/keys
Groq	Yes	Yes (whisper-large-v3)	None — pick after checking the key	console.groq.com/keys
Custom (OpenAI-compatible server)	Yes	Yes (whisper-1)	None	Your server

Each card has a Get a key link that opens the provider’s key page.

WordPress AI (site connection)

On WordPress 7.0 and later, a site owner can connect an AI provider once for every plugin under Settings → Connectors. When that is available, SimplifyGuide lists WordPress AI (site connection) first and makes it the default, so there is no separate key to paste. The card shows Connected or Not connected, a Test connection button, and a link to Manage connections in Settings → Connectors.

On older WordPress versions, or where AI has been switched off for the site, this option does not appear and the default provider is Anthropic.

Anthropic, OpenAI, DeepSeek, OpenRouter and Groq

Each is a key you create in the provider’s own console and paste here. You pay the provider directly for what you use. Anthropic uses its own API; the others all speak the OpenAI-compatible protocol.

Custom (OpenAI-compatible server)

For anything else that speaks the OpenAI API: a local Ollama, LM Studio or vLLM server, a hosted model gateway, or a company proxy. Set Server URL (for example http://localhost:11434/v1 for Ollama), a model name, and a key if your server needs one.

“Local” means local to your WordPress server, since requests are made by the server, not your browser. A model running on your laptop is only reachable if WordPress runs there too.

Setting up a provider

1. Choose it in Provider at the top of the tab.
2. Open its card in the Providers list and paste the key into API key.
3. Press Check key. SimplifyGuide lists the models the key can use, then makes one tiny test request — the only reliable way to learn whether the account still has credits.
4. Pick a Model. After a successful check the field suggests every model the key can use; if it was empty, a sensible chat model is chosen for you.
5. Press Save changes.

A successful check reads Key works. N models available; test request to … succeeded. You can check a key before saving it.

Other fields on the card:

Field	What it is for
Show	Reveals the saved key. Only appears once a key is saved.
Model	The model that writes step text. Any model ID the provider accepts.
Transcription model	Voice-capable providers only (OpenAI, Groq, Custom).
Use your own server (optional)	Point this provider at a different OpenAI-compatible endpoint, such as a proxy. Leave empty for the provider’s own address. On the Custom card this field is Server URL and is required.
Remove saved key on save	Deletes the stored key when you press Save changes.

Leaving API key blank when you save keeps the key you already stored. A saved key is never printed back into the page; the field shows only its last four characters.

Voice transcription

Transcription uses its own provider, set under Settings → Voice narration → Transcribe with: OpenAI (the default), Groq or Custom. It uses the key from that provider’s card in AI writing → Providers, so you can write with Anthropic and transcribe with OpenAI, for example.

Narration files up to 25 MB are accepted, the common limit of transcription APIs. See Voice narration.

What is sent

Requests go from your server straight to the provider you chose. Your browser never talks to the provider and never sees the key.

Feature	Sent to the provider
Writing step descriptions	The guide title and, per step: its action, current text, element label, page path, and a little page context (element type, nearest section heading, page heading, page title, field help text)
Rewriting one piece of text	That text only, up to 4,000 characters
Transcribing narration	The narration audio file
Check key / Test connection	A model list request and a one-word test prompt

Never sent: screenshots, narration (unless you are transcribing), passwords, anything Smart Blur detected, or the value of an element you blurred. Step text can quote an ordinary value you typed — Type “Summer sale” in “Title” — and that text is what the model rewrites, so it goes too. The instructions given to the model also tell it never to include passwords or personal data. The provider’s own terms and privacy policy apply to what it receives.

Keys are encrypted at rest

API keys are stored in the simplifyguide_ai option, encrypted with AES-256 using a secret derived from your site’s AUTH_KEY salt. A database dump on its own does not reveal them.

Two consequences:

• If you change your site’s salts, stored keys can no longer be decrypted. Paste them again.
• If PHP’s OpenSSL extension is missing, keys are stored base64-encoded instead — obscured, not encrypted. Almost every host has OpenSSL.

Who can use AI

AI calls cost the site owner money, so the bar is higher than for recording.

Action	Needs
Rewrite steps, transcribe narration	publish_posts (Authors and above)
Change providers, check or reveal keys	manage_options (Administrators)

Users who can record but not publish — Contributors, by default — still get rule-based step text; the editor treats AI as unavailable for them, and their narration is saved as audio without being transcribed. Developers can change the capability with the simplifyguide_ai_capability filter.

Rate limit

Each user can make 60 AI requests per hour. A request is one rewrite, one “describe these steps” call (however many steps it covers), or one transcription. Past the limit, the user sees You have reached the hourly limit for AI requests. Try again in N minutes.

Change or remove the limit with the simplifyguide_ai_rate_limit filter; zero or less turns it off.

Error messages

SimplifyGuide turns provider errors into a sentence you can act on:

Message	What to do
… rejected the API key. Check that it is correct and active.	The key is wrong, revoked or for another account. Paste a fresh one.
Your … account has no credits left. Add credits or use another provider.	Top up with the provider, or switch provider.
The model ”…” was not found at …	Press Check key and pick a model from the list.
… is rate-limiting requests. Wait a minute and try again.	The provider’s own limit, not SimplifyGuide’s.
… had a server error (5xx). Try again shortly.	The provider is having trouble.
Could not reach …	Your server could not connect: firewall, DNS, or a wrong Server URL.
… answered, but not in the expected format.	Try again, or choose a larger model. Small local models sometimes fail here.

See Troubleshooting for more.