overlay: 1.0.0 info: title: API Evangelist enhancements for the Furiosa Model Server Model Repository API version: 1.0.0 x-generated: '2026-08-16' x-method: generated x-source: openapi/furiosa-model-repository-v2.yaml x-note: >- Non-destructive enhancements to the spec FuriosaAI publishes at python/furiosa-server/openapi/model_repository.yaml. The original is saved verbatim. The material addition here is a security posture note: every operation on this surface changes server state and none declares any authentication. extends: openapi/furiosa-model-repository-v2.yaml actions: - target: $.info update: x-provider: FuriosaAI, Inc. x-provider-url: https://furiosa.ai/ x-product: Furiosa Model Server (furiosa-server) x-spec-origin: >- Vendored by FuriosaAI into furiosa-ai/furiosa-sdk as the published contract of Furiosa Model Server's model-management surface. info.title/description name the upstream Triton Inference Server Model Repository extension because that protocol is implemented verbatim. x-deployment: customer-hosted x-security-posture: >- No securitySchemes and no security requirement are declared, and the furiosa-server README lists authentication as an unchecked roadmap item. Load and unload of arbitrary models from the repository root are therefore unauthenticated - this surface must not be exposed beyond a trusted network. - target: $.servers update: - url: http://{host}:{port} description: Customer-hosted Furiosa Model Server REST endpoint (default port 8080). variables: host: default: localhost port: default: '8080' - target: $.paths['/v2/repository/index'].post update: x-idempotent: true x-note: >- A POST that reads. Modelled as POST by the Triton extension because it takes a body ({"ready": bool}); it has no side effects. - target: $.paths['/v2/repository/models/${MODEL_NAME}/load'].post update: x-idempotent: false x-consequence: state-change x-response-semantics: Bare 200 on success, 400 on failure, no body (BooleanResponse). - target: $.paths['/v2/repository/models/${MODEL_NAME}/unload'].post update: x-idempotent: false x-consequence: state-change x-response-semantics: Bare 200 on success, 400 on failure, no body. x-warning: Unloading a model in use will fail in-flight inference requests against it.