AI providers
Configure the provider keys the Assistant and Studio use.
Settings > AI Providers is where you connect the language models your workspace's roles and Studio use. Only workspace admins can reach it.
Providers
| Provider | What it needs | Notes |
|---|---|---|
| Skilder | Nothing | Platform-managed. No key to configure. Usage counts against your daily budget, see below. |
| Infomaniak AI | API key, Product ID | |
| Infomaniak AI (v1) | API key, Product ID | Older model set. Kept for workspaces already using it. |
| OpenAI | API key | |
| Anthropic | API key | |
| Google Gemini | API key | |
| Amazon Bedrock | AWS region, and an AWS access key or a Bedrock API key | Runs on your own AWS account, see below. |
| Exoscale | API key | Open models hosted in Vienna (Austria). Private preview, see below. |
| Ollama | Base URL | Local models. Not available in production deployments. |
You can also add a custom provider: an OpenAI-, Anthropic-, Google- or Ollama-compatible endpoint you point at yourself (bring your own key, or BYOK).
Configure a provider
For a catalog provider, click the provider card:
- Enter its API Key (and Product ID for Infomaniak, or the Ollama URL for Ollama). The first time you configure a provider by key, you also acknowledge its usage terms.
- Click Next: Select Model, then choose which of its models the workspace can use. Starred models are recommended.
- Click Save.
Skilder needs no key: pick its models and save.
Skilder models
Each Skilder model is named with the company that hosts it and the country it runs in:
| Model | Selected in a new workspace |
|---|---|
| Qwen3.8 — Infomaniak (Switzerland) | Yes, and the default model |
| Apertus 1.5 — Infomaniak (Switzerland) | Yes |
| Mistral Small 4 — Infomaniak (Switzerland) | Yes |
| Kimi K2.6 — Infomaniak (Switzerland) | Yes |
| Qwen3.8 — Exoscale (Austria) | No: an admin selects it on the Skilder card. Hosted in Vienna. |
| Claude Sonnet 4.6 — Amazon Bedrock (EU) | No: an admin selects it on the Skilder card. Every call starts in an AWS European region (EU, Switzerland or UK) and AWS routes it within its EU geography, never to the US or a global region. |
Qwen3.8 is offered by two hosts as two separate models; one does not stand in for the other. Exoscale's service is in private preview, and its price may change. Claude Sonnet 4.6 is charged at Amazon's published Bedrock price for Europe (Zurich); other European regions may price differently. It costs five to six times as much per output token as Qwen3.8, so the daily budget runs out sooner when it is used.
For a custom provider, click Add custom provider and fill in API Type, API URL, API Key, and a comma-separated list of Available Models.
Amazon Bedrock
Amazon Bedrock connects to your company's AWS account, so usage is billed by AWS and stays in that account.
- Choose the AWS region Skilder calls Bedrock in.
- Choose a sign-in method and enter its credentials:
- AWS access key: the access key ID and secret access key of an IAM user. To replace one, enter both values.
- Bedrock API key: a single key created in the Amazon Bedrock console.
- Click Next: Select Model. The list shows the models your account can use in that region.
Give the credentials only what Skilder needs: bedrock:ListFoundationModels, bedrock:ListInferenceProfiles, bedrock:GetFoundationModelAvailability, bedrock:InvokeModel and bedrock:InvokeModelWithResponseStream. Without bedrock:GetFoundationModelAvailability, the list also includes models your account may not be able to use.
| What you see | What to fix |
|---|---|
| AWS refused these credentials | Check the keys are correct and active, and that their policy allows the two List actions above. |
| No text model available | Enable model access in the Amazon Bedrock console, or choose another region. |
| A model returns 403 when someone uses it | Enable that model in the Amazon Bedrock console, or allow the two Invoke actions above for it. |
Exoscale
Exoscale on-demand inference runs open models, such as Qwen3.8-27B, in Exoscale's Vienna (Austria) zone. Create an AI API key in the Exoscale console and enter it as the API Key. The list shows the chat models that key can use; the embedding model is not offered.
The service is in private preview: no SLA, shared capacity and no published rate limits, and models and limits may change. Skilder names Exoscale in the chat error when a call is refused:
| What you see | What to fix |
|---|---|
| 401 or 403 | Check the key is correct and active, and that it is allowed to use the chosen model. |
| 429 or 503 | Exoscale is rate limiting or out of capacity. Try again shortly. |
Remove a provider
Click the trash icon on a configured provider card and confirm.
Removing a provider deletes its key and all its settings. This cannot be undone. The platform-managed Skilder provider cannot be removed.
Default model
Default workspace model picks which configured model is used when a role doesn't specify one.
Daily budget
Skilder-provider usage is capped by a daily credit budget. The default cap is 1 credit per day. It resets at midnight in the workspace's timezone. BYOK and custom providers are never capped.
Everything that runs on the Skilder provider draws on the budget: chat, skill suggestions for a role, AI help with an MCP server configuration, and the sub-agents a skill script delegates to. Once the budget is spent, each of them stops with a message that the workspace has reached its daily AI usage limit.
There is no admin control to raise the cap yourself: View usage details on the Skilder card only shows this month's token consumption. Add your own provider key to stop drawing on the shared budget.
Security
- API keys, AWS access keys and product IDs are encrypted at rest. A saved key is never shown again, even to an admin.
- Only admins can configure, edit or remove a provider. Contributors and members can read the model catalog to use in chat, but cannot see or change keys.
- Per-provider usage, including a per-user breakdown, is visible to admins only.

