Skip to main content
The hiddenlayer-apim CLI deploys HiddenLayer policy fragments to Azure API Management (APIM) and applies those fragments to APIs that proxy LLM traffic. Use the CLI when you want a repeatable way to add Runtime Security guardrails without hand-editing large APIM policy XML blocks. The CLI can protect requests before they reach the model, responses before they return to users, or both. It supports package-based deployment so you can choose the evaluation flow that matches your architecture. Deploy v2-request-evals for inbound-only protection, v2-response-evals for outbound-only protection, or both packages for full request and response guardrails.

​
How It Works

  1. hiddenlayer-apim deploy creates HiddenLayer named values and policy fragments in your APIM instance.
  2. hiddenlayer-apim apply <api-id> reads the target API policy and injects <include-fragment> references.
  3. APIM calls HiddenLayer evaluation endpoints during inbound and/or outbound processing.
  4. HiddenLayer returns a provider-shaped payload. APIM forwards the payload, redacted payload, or block response based on the policy decision.

​
Fragment Packages

Recommended pathUse the v2 packages for new APIM guardrails deployments. The v2 endpoints accept provider-native payloads and return provider-shaped payloads, which keeps the APIM policy simpler.

​
Provider Compatibility

Provider support depends on which package you deploy:
  • v2-request-evals and v2-response-evals forward the original provider request and response bodies to HiddenLayer’s v2 evaluation APIs and use whatever HiddenLayer returns. The APIM fragments do not parse provider-specific fields. Provider support is therefore whatever HiddenLayer’s /detection/v2/request-evaluations and /detection/v2/response-evaluations endpoints accept (OpenAI, Azure OpenAI, Anthropic Messages, and additional providers as they are added). Use the HL-Provider-Id request header to override auto-detection and HL-Runtime-Edge-Provider to label the edge integration.
  • v1-interactions is OpenAI chat-completions shaped. It reads OpenAI request fields (model, messages) and OpenAI response fields (choices[].message), and its block response is an OpenAI chat-completion payload. Use it for Azure OpenAI / OpenAI chat-completions APIs.
See Fragment packages for the full per-package breakdown.