Models and providers

Provider routing, model selection, account availability, and model capability are separate facts in Iteron. A model can appear in a catalog and still be disabled because the provider is disabled, no credential is present, or account entitlement is unknown.

Resolve a route

Routing-sensitive provider precedence is:

1. CLI --provider;
2. ITERON_PROVIDER;
3. trusted user config ~/.iteron/config.json;
4. built-in default glm.

Model selection accepts --model, ITERON_MODEL, trusted user config, and one repository-safe exception: .iteron/config.json may name a bare model within the already trusted provider. It cannot qualify another provider or redirect egress.

Use provider:model where qualification is needed and the provider id is known:

iteron --model anthropic:PROVIDER_MODEL_ID

Inspect availability

In the TUI:

• /model shows the current selection and model picker;
• /model retry MODEL_ID explicitly retries one unavailable catalog path;
• grey or disabled entries are not selectable;
• /status reports the resolved route and effort application.

Iteron does not infer that a model is billable merely because its id is documented. Provider errors and catalog evidence may change availability during a session.

Compatible endpoints

--base-url and ITERON_BASE_URL create a trusted one-run OpenAI-compatible Chat endpoint override. Persistent endpoints belong in user configuration. The root must include its full path/version prefix and use HTTPS except for an exact loopback host.

OpenAI-compatible wire shape does not make provider business error codes interchangeable. User-defined instances should use the correct error_profile or the conservative custom profile.

See configure a provider and the provider matrix.