Settings and billing

AI models and your own key

Walden calls an AI model for chat, new ideas, the second reader and UI sketches. Settings → AI shows which model each one starts with, and is where an owner can add the workspace’s own provider key.

The model each part starts with

Open Settings and choose AI under This workspace. There’s a row for each part of Walden whose model you can choose:

Chat and new ideas: Chat beside an idea, and the first draft of an idea from Home. One choice covers both.
Second reader: The readers who leave questions in the margin.
Sketches: Drawing UI sketches, one place at a time.

An owner picks each one from the models that can run in the workspace. Everyone else sees which it is. Until an owner picks, Walden’s own choice stands. On Free, chat starts on a model that costs less, so the day’s credits go further.

Pick your own

Anyone can pick another model where they use it: by the model’s name beside the chat box, under the box on Home, in the second reader, and under a flow beside Sketch the flow. The list shows roughly what one call costs on each model.

Your pick is yours alone, and follows you to every workspace where that model can run.
Pick the workspace’s own model again, and you follow its default from then on.
A stronger model uses more credits for the same job. See Plans and credits.

A few calls are always Walden’s pick, whatever is chosen for sketches: the check a sketch gets before you see it, and a change to a sketch asked in chat. If the model you picked isn’t being served just then, the next pick answers instead. If it simply can’t be reached, Walden says so and you try again.

Use your own provider key

If your workspace already pays a model provider, Walden can send its calls with that key.

In Settings → AI, go to Your own model key.
Choose the provider: Anthropic, OpenAI or Google.
Paste the key and choose Store key.

Walden checks the key with its provider before keeping it, and keeps it encrypted. From then on:

Calls go under your account and your terms with that provider, and use no credits.
Only that provider’s models are offered, with no prices beside them.
Settings shows the provider and the key’s last four characters.

If the key stops working, Settings says so beside it, and calls are charged in credits until a working key is stored. Remove takes the key off at once, and credits apply again. The key is also removed when the paid plan ends.

Your words and training

Walden asks for every call to go only to providers that have agreed not to train on what is sent, and Walden doesn’t train models on what you write either. With your own key, requests go under your account and your terms with that provider. The privacy page says this in full.