Sign in
Internal

Edit a MANAGED backend's knobs (platform admin)

Weight, health kill switch, concurrency cap, and prompt-byte lane on one managed backend. Identity fields (provider, endpoint) and self-hosted pool:// backends are refused here: pool state belongs to the serving deployments that own it (their /weight and /health endpoints).

patch/v1/admin/inference/catalog/{modelId}/backends/{backendEndpoint}

Path parameters

modelIdstring required
backendEndpointstring required

Request body

healthyboolean
maxConcurrencyinteger
maxPromptBytesinteger
minPromptBytesinteger
reasonstring

Audit note, required when flipping the kill switch

weightinteger

Response

Applied

enabledboolean
modelIdstring required
statusstring
updatedAtstring

Changes