Skip to main content

AI Prompt Guard Configuration

The ai-prompt-guard plugin safeguards prompts to LLM using allow/deny patterns, ensuring only approved inputs pass. It can check the latest message or full history. Configuration covers plugin parameters.

Parameters​

See plugin common configurations for configuration options available to all plugins.

  • match_all_roles

    boolean

    default: false


    If true, validate messages from all roles. If false, validate the message from user role only.

  • match_all_conversation_history

    boolean

    default: false


    If true, concatenate and check all messages in the conversation history. If false, first select only the final message, then apply match_all_roles. With the default role setting, the final message is checked only when its role is user; the plugin does not search backward for an earlier user message.

    This option does not apply to Responses API requests because instructions and input are parallel fields rather than conversation history.

  • allow_patterns

    array[string]

    default: []


    An array of RegEx patterns that selected text should match. When configured, the combined selected text must match at least one pattern to be considered valid.

    In APISIX 3.20.0, string items in an Embeddings input array are joined with spaces and evaluated as one value. A pattern can match within an item or across the inserted space. API7 Enterprise 3.9.23 and 3.10.10 do not inspect Embeddings string arrays.

  • deny_patterns

    array[string]

    default: []


    An array of RegEx patterns that messages should not match. If messages match any of the patterns, the request should be rejected.

    If both allow_patterns and deny_patterns are configured, the plugin first ensures that at least one allow_patterns is matched.

    In APISIX 3.20.0, deny patterns run against the same space-joined Embeddings value as allow patterns. A match anywhere in that combined text rejects the whole batch. API7 Enterprise 3.9.23 and 3.10.10 do not inspect Embeddings string arrays.

  • fail_mode

    string

    default: skip

    valid value:

    skip, warn, or error

    Behavior when the plugin receives a request it cannot inspect, such as non-AI traffic on a Consumer binding. With skip, the request passes unchecked. With warn, it passes unchecked and a warning is logged. With error, the plugin rejects it with HTTP 400. None of these outcomes means prompt validation succeeded.

    Introduced in API7 Enterprise 3.9.14 and APISIX 3.18.0.

  • max_req_body_size

    integer

    default: 67108864

    valid value:

    greater than or equal to 1

    Maximum request body size in bytes buffered into memory. Larger request bodies are rejected with HTTP 400 before prompt validation. Introduced in API7 Enterprise 3.9.17 and 3.10.4, and APISIX 3.18.0.