Run the model
Compile a reviewed serving plan for infrastructure you control, then deploy through the CLI, API, SDK, or Terraform.
Run InferCrane →Hugging Face to stable API
Choose a model. Get a production endpoint.
Start with a Hugging Face repository. InferCrane turns it into an inspectable serving plan before anything runs.
Plan a model →Find a model
Choose a current repository or paste the exact Hugging Face ID.
Discovery metadata only. Your plan determines runtime, GPU fit, cost, and measured performance.
ModelTaskUpdatedSignalsAction
One endpoint contract
Choose where inference runs without coupling the app.
Connect compatible capacity
Put an existing OpenAI-compatible endpoint behind the same model identity, privacy acknowledgement, request limit, and cost ceiling.
Review the connection flow →Use qualified managed capacity
A callable route appears only when supplier availability, public pricing, funding, and usage reconciliation are current. Otherwise the catalog remains fail-closed.
Check current availability →Exact model identity
The catalog is a shortcut, not an allowlist.
Submit another Hugging Face repository, an immutable custom OCI workload, or an endpoint your team already operates. Compatibility is checked in the plan before any infrastructure mutation.
infercrane modelsinfercrane workload init ./my-model --model organization/repositoryinfercrane workload plan ./my-model