Connect a Chat Model
Step-by-step guide to connecting a chat model in Sciwand - provider steps vs Sciwand steps, billing, and vision models.
New to this? Start with Connecting AI Providers for what an API key is and which model to pick.
Chat models power the Sciwand AI assistant. Connecting one has two halves: a few steps at the provider (create a key, add a little credit) and a few steps in Sciwand (paste the key, enable models).
No subscription needed. API access is pay-as-you-go and is billed separately from any consumer plan (such as ChatGPT Plus or Claude Pro). You pay only for what you use - often just cents. See Connecting AI Providers for details.
Part A - at the provider
- Sign in to the provider developer console (create a free account if you do not have one).
- Open the API keys page, create a new key, and copy it. Keep it somewhere safe - most providers show it only once.
- Open Billing and add a small amount of credit (or a payment method). Without a balance the API usually returns a "quota" or "billing" error. Some providers have a free tier.
Part B - in Sciwand
- Open Settings and select the AI Models tab.
- Click Manage (or Add Model, if you have none yet) to open Configure Models.
- Pick your provider from the list at the top.
- Click Configure to open the provider dialog.
- Paste your key into API Key (the Get your {provider} API key link opens the right page). For a local provider, enter the Server URL instead.
- Click Save Changes.
- Sciwand fetches the available models - tick the checkbox next to each model you want to enable.
- (Optional) Mark one model as the Default; it is used unless you switch models inside a chat.
When you enable a model, Sciwand runs a quick background smoke test. If the key or model is not valid, the checkbox reverts and an error is shown.
Vision models
Some tasks (reading figures, scanned pages) use a vision-capable model. You can pick a separate Vision Model; if you leave it unset, your chat model is used. Vision is supported on Claude, Gemini, OpenAI, and OpenAI-compatible providers.