DocsReference
Models and your own keys
Pick the model that runs your benchmark, on our credentials or your own.
Sioma serves context. It does not need a model of yours to do that, and resolution never spends your provider credits.
The model you pick here runs your benchmark, and the question generation behind it. That is what this page is about.
You manage it in Control room, under Configure. A workspace can connect several providers at once. Exactly one is active, and the active one runs your next benchmark.
Managed models#
The simplest option: pick a plain model name and it runs on our credentials. Which cloud serves it is a routing decision on our side.
Managed runs count against your workspace's free run allowance. Your dashboard shows what is left in the current window.
Bring your own key#
Connect your own provider account and your runs go on your own bill, with no allowance to run out. Each provider asks for its own native credentials:
| Provider | What it asks for |
|---|---|
| Anthropic | API key |
| OpenAI | API key |
| Azure | Resource name and API key |
| Amazon Bedrock | Region, access key id, secret access key |
| Google Vertex | Project id, location, service-account JSON |
Connecting verifies the credentials by listing that provider's models, and you then choose from that live list. You never type a model id, so a run cannot fail on a name that does not exist in your account.
Your own connection also widens what you can pick: a few premium models are available only on your own key.
How your credentials are stored#
Encrypted with AES-256-GCM, and never stored in plaintext. Reading a connection back gives you a non-secret hint, never the value. If the encryption key is rotated or lost, the connection reads as having no credentials and the run fails cleanly, rather than quietly falling back to another provider.
Your own endpoint#
Two more options for teams that host their own:
- Custom — any OpenAI-compatible endpoint. Give a base URL and an optional key. This covers a self-hosted gateway, or a hosted Ollama.
- Local — Ollama, or the Claude Code CLI on your own machine. No credentials, and intended for local development.
What is metered#
Managed runs are metered against your allowance, since they spend our credentials. Runs on your own key, a custom endpoint, or a local model are your own bill and are never metered.