Schema Enforcement

A model adds the type or schema structure a Guardrail needs to parse, validate, and, for schema-backed formats, transform a payload. Zilla accepts a model wherever a binding validates part of a request or response, from a Kafka message key to an HTTP request body.

Primitive Models

Primitive models validate a value against a basic type, with a few type-specific properties:

model: string
encoding: utf_8
maxLength: 256

• Boolean
• Double
• Float
• Int32
• Int64
• String

Schema Models

Schema-backed models reference a catalog to supply the schema used to validate, and optionally transform, the payload. A schema can be resolved by subject, by topic, or by a specific id:

model: json
catalog:
  my_catalog:
    - subject: tool-payload

• Avro
• JSON
• Protobuf

The Avro and Protobuf models also accept a view, which transforms the payload between the wire format and the schema format. That transform is what Transform covers.

Guardrails in Practice

The http binding accepts a model on request and response content, headers, path, and query parameters, so a Guardrail can reject a malformed tool call or upstream response before it reaches an agent:

options:
  requests:
    - path: /tools/call
      method: POST
      content:
        model: json
      responses:
        - status: 200
          content:
            model: json

See the http server and http client reference for the full set of request and response validation options, which currently cover the primitive and schema-backed model names above. Kafka Gateway's Validate shows the same schema-backed model paired with a catalog to validate against a schema pulled from a registry; see Validate for how that pairing applies as a Guardrail.