Guardrails Per Model

Runs body-inspecting guardrails only for a model selected from the request body

Versions marked “overview” do not contain this page. Selecting one opens that version’s documentation overview.

Category: Setup-dependent integration
Task: Runs body-inspecting guardrails only for a model selected from the request body

Prerequisites: The external service, credentials, or certificates referenced by this configuration.

Run it: Use ghcr.io/praxis-proxy/praxis:0.7.2 and follow the first reverse-proxy tutorial to mount and start the configuration.

This configuration comes from the selected release. The example has not been run here; external services are not bundled.

Download the source file.

# Per-Model Guardrails
#
# Runs body-inspecting guardrails only for a model selected from the request body.
#
# `json_body_field` promotes the JSON `model` field into the reserved
# `x-praxis-guard-model` header during StreamBuffer pre-read, and the
# `guardrails` filter is gated on that header so it inspects bodies only for
# the guarded model. Other models skip guardrails entirely.
#
# The gate header must be a reserved `x-praxis-*` name: clients cannot supply
# reserved headers (they are rejected at ingress), so the gate reflects the
# promoted body value and cannot be spoofed. Conditionally running a security
# filter is a deliberate bypass for non-matching requests, so it is opted into
# explicitly via `insecure_options.skip_pipeline_checks.conditional_security`.
#
listeners:
  - name: default
    address: "127.0.0.1:8080"
    filter_chains:
      - main

filter_chains:
  - name: main
    filters:
      # Promote the body's `model` field into a reserved, client-unspoofable
      # header during pre-read.
      - filter: json_body_field
        field: model
        header: x-praxis-guard-model

      # Inspect bodies only when the promoted model is the guarded one.
      - filter: guardrails
        conditions:
          - when:
              headers:
                x-praxis-guard-model: guarded-model
        rules:
          - target: body
            contains: "forbidden"

      - filter: router
        routes:
          - path_prefix: "/"
            cluster: backend

      - filter: load_balancer
        clusters:
          - name: backend
            endpoints:
              - "127.0.0.1:3000"

insecure_options:
  allow_private_endpoints: true # example proxies to a local backend
  skip_pipeline_checks:
    # Per-model guardrails intentionally bypasses guardrails for non-guarded
    # models; acknowledge the conditional-security bypass explicitly.
    conditional_security: true