Models and lanes
Select Strix by intent rather than binding your integration to a provider model.
Canonical public IDs use the strix-* prefix. Start with strix-auto unless the task already has a clear intent.
| Model | Intended workload |
|---|---|
strix-auto | Classify the request and select a lane |
strix-coding | Coding and tool-use loops |
strix-quick | Fast mode for small, focused, reversible tasks |
strix-deep | Complex multi-file work and deep reasoning |
strix-ultrabrain | Maximum reasoning for the hardest tasks |
strix-bug-hunt | Reproduction, debugging, and root cause |
strix-refactor | Behavior-preserving refactoring |
strix-arch | Architecture analysis and design review |
strix-visual | UI, CSS, and visual engineering |
strix-writing | Documentation and polished prose |
strix-research | Evidence-heavy investigation |
strix-sonnet | Baseline comparison lane |
GET /v1/models is the authenticated source of truth for the currently exposed catalog. Legacy aliases can remain compatible, but new integrations should send canonical strix-* IDs.
A lane is an intent contract, not a promise of a fixed upstream provider. Routing may change while the public model ID stays stable. Read x-strix-lane and related completion headers when you need evidence about a particular request.