Models and routing overview
Your agents can use models from many LLM providers without a separate account or API key for each one. Your agent names a model, and Alfe routes the request to the provider that serves it. Unless you bring your own credentials, usage is billed from your credit pool.
Supported providers
Section titled “Supported providers”The catalog covers these providers:
- Direct providers — Anthropic, OpenAI, Google, DeepSeek, MiniMax, Mistral, xAI, Zhipu AI, Moonshot AI, and Qwen.
- OpenRouter — additional models from many other providers.
The catalog changes as models launch and retire, so this page doesn’t list individual models. The picker in the dashboard always shows what’s available to you. It reflects your organization’s policy and your agent’s runtime.
Why routing through one place helps
Section titled “Why routing through one place helps”Because requests share a single route, you get a few things for free:
- Switching stays simple. Changing an agent’s model doesn’t touch its channels, integrations, or workspace files.
- Usage lands in one place. The Usage page shows spend across providers, so you don’t reconcile separate bills.
- Policy is enforced centrally. When an administrator restricts models, the restriction applies to every agent.
Where the runtime fits in
Section titled “Where the runtime fits in”Some runtimes run any model in the catalog. Others are tied to one provider’s models: a Claude Code agent uses Anthropic models, and a Codex agent uses OpenAI models. The picker only offers models the agent’s runtime can use, and it says why when the list is locked.
Next steps
Section titled “Next steps”- Choose a default model — pick a model by use case, and see what happens when one retires.
- Bring your own key or subscription — run requests on your own provider account.
- How model usage is billed — how each kind of credential is paid for.