Skip to main content
Once you’ve configured one or more model providers, you can choose which one Cyrus uses by default for every Linear issue and agent run.

​
Change the default model provider

1

Go to Behaviours settings

Navigate to app.atcyrus.com/settings/behaviours.
2

Select a provider

Find the Default Model Provider dropdown and select the provider you want Cyrus to use by default (Claude, OpenAI/Codex, Gemini, or Cursor).
This setting applies globally — every new issue and agent run will use the selected provider unless overridden per-issue. On the same page, use Default Claude Model or Default Codex Model to choose the model for that provider. The built-in defaults are opus for Claude and gpt-5.5 for Codex. Available choices depend on your configured provider and authentication method; see OpenAI model availability before choosing a Codex default.

​
Override the provider per issue

You can override the default provider for a specific issue in two ways:

​
Using an inline tag in the issue description

Add one of the following tags anywhere in your Linear issue description:

​
Using a Linear label

Apply the provider name as a label on the issue (openai is also accepted for Codex):

​
Override the model per issue

You can also override the specific model (not just the provider) for a given issue, either with a [model=...] description tag or a Linear label of the same name.

​
Supported models by provider

Team or runtime configuration can override these defaults. Selecting a model requires access through the provider and a compatible Cyrus runtime. The linked provider pages list current choices, version requirements, and availability limitations.

​
Gemini models

The aliases above are expanded for bare Linear labels. Use the full model ID in description tags or provider/model labels, for example [model=gemini-3-pro-preview] or gemini/gemini-3-pro-preview. Other Gemini IDs can be passed the same way if your runtime and Google account support them.

​
Select a provider and model together

A single Linear label can specify both:
Each line is a separate example. For an inline override, use:
Or select a model with a bare label such as fable, opus, gpt-6-astra, or gemini-2.5-flash. Full Claude model IDs need a [model=...] tag or a claude/ label prefix.

​
Selection priority

  • Description tags take priority over labels for the corresponding provider or model selection.
  • A combined provider/model label takes priority over separate provider and model labels.
  • An explicit provider takes priority over the provider inferred from a model name. If they conflict, Cyrus uses the explicit provider’s default model.
  • With no explicit provider, recognized Claude, GPT, and Gemini model names infer the provider. Otherwise Cyrus uses the configured default provider.
Label and provider-name matching is case-insensitive. Apply one provider/model choice per issue to avoid conflicting selectors. See Labels and Routing for examples alongside workflow labels.

​
OpenCode models on self-hosted runtimes

OpenCode is available as a self-hosted runner; it is not a provider choice in the hosted dashboard. Install and authenticate OpenCode on the machine running Cyrus, then use an explicit agent tag with an OpenCode provider/model ID:
The equivalent Linear label is opencode/openai/gpt-5.5. Use a model available through your configured OpenCode provider. On fully self-managed installations, opencodeDefaultModel and opencodeDefaultFallbackModel in ~/.cyrus/config.json set the runtime defaults. Cyrus does not select a built-in OpenCode model. Explicitly select OpenCode: a bare [model=openai/gpt-5.5] only infers OpenCode when inferOpenCodeRunnerFromProviderModel is enabled. See the Cyrus configuration reference for provider and runtime configuration.

​
Agent sessions

The agent (provider) cannot be switched within an active agent session. If you need to use a different provider on the same issue, a new agent session will be spawned with the selected agent.
cyrus-f1

Cyrus Community

Get support on Discord