PrismicalPrismical

AI Models & BYOK

Pick the models behind transcription and text generation, or connect your own provider keys.

Everything AI in Prismical runs on a model, and by default that's a model we manage for you — nothing to set up, no key to paste, no bill from anyone but us. This page is for when you want your own.

Auto, and Why You Might Leave It Alone

Out of the box every AI feature uses Prismical Cloud · Auto. We pick the model, we pay for it, and we swap in better ones as they ship. For most people this is the right setting forever.

Bring your own key when you want a specific model, your own provider bill, or your own data-processing agreement with that provider.

Where the Settings Live

Settings → AI Models has two defaults:

DefaultWhat it powers
TranscriptionTurning speech into text — recordings and anything else Prismical captures. See Transcription models.
Text generationEnhance, skills, and other text written from your notes.

Ask AI is deliberately not here — you pick its model per question, in the Ask panel itself.

Connecting Your Own Provider

Premium

Bringing your own AI keys is a Premium feature.

See what each plan includes.

Prismical works with OpenAI and OpenRouter.

Settings → AI Models, with the managed defaults up top and a grid of providers to add below.

Go to Settings → AI Models and click Add a provider.

Pick OpenAI or OpenRouter.

Give it a Label — something like "Personal OpenAI". This is just for you; it's how the provider shows up in model pickers.

Paste your API key and click Connect. Prismical fetches your available models to check the key works.

Walk through the rest of the wizard to choose which models to use for Ask AI, skills, and — for OpenAI — transcription.

The provider then appears under Connected, and its models become available in the pickers.

OpenRouter does text only. If you want to transcribe with your own key too, that needs OpenAI.

Which Models You'll See

Prismical asks your provider what your key can access rather than keeping a list of approved models, so new models generally show up on their own. What comes back is sorted into text models and transcription models, and anything that's neither — image generation, text-to-speech, moderation — is left out.

Choosing Which Models Appear in Ask AI

Ask AI's picker is curated, so it doesn't become a wall of every model your key can reach. Open a connected provider and use Models in Ask to tick the ones you want.

Leave every model unticked and that provider won't appear in the Ask picker at all. If you connected a key and Ask still only offers Auto, this is why.

Changing a Default

Click Change model on either default card. Pick the source — Prismical Cloud or one of your connected providers — then the specific model.

Skills and transcription each use one model for everything. There's no per-skill model choice.

If a Key Stops Working

Where you find out depends on where you are.

In Settings, the model list is fetched live, so a bad key shows up as "Couldn't load models — check the API key, then retry."

In Ask AI, the key isn't checked until the answer is already being generated, so a bad key surfaces as a generic "Something went wrong."

If Ask AI keeps failing on your own model and you can't tell why, go to Settings → AI Models, open the provider, and look at the model list. That's where the real error is. Switching back to Auto confirms it in one question.

If you delete a connected provider, anything pointing at it quietly falls back to Auto rather than breaking.

How Your Keys Are Stored

Your API keys are encrypted before they're stored and are never sent back to your devices as part of normal sync — Prismical decrypts them server-side, only when a request of yours needs them.

Each key is encrypted with its own key, which is itself encrypted with a master key held outside the database. The encryption is tied to the row it belongs to, so a copied ciphertext won't decrypt anywhere else. See Security & privacy.

Questions

On this page