Agent Schema

The open, vendor-neutral schema for describing AI agents and making them interoperable across tools, runtimes, and marketplaces.

v1.0.0 · JSON Schema Draft-07

Canonical format for agent metadata

Agent Schema defines a single, machine-verifiable format for AI agent metadata — identity, capabilities, deployment details, costs, and safety-related information — so different ecosystems can speak a common language about agents.

The schema is maintained in the minuetai/agents repository and is designed to sit alongside existing protocols and frameworks, not replace them.

View full schema.json Raw JSON (latest)

Excerpt from the Agent Schema (agent-specific surface):

{
  "$schema": "http://json-schema.org/draft-07/schema#",
  "$id": "https://minuet.dev/schema/agent-schema/v1.0.0",
  "title": "Agent Schema",
  "description": "Canonical schema for AI agent metadata and interoperability.",
  "type": "object",
  "required": [
    "agent_id",
    "name",
    "description",
    "model_lineage",
    "endpoint",
    "capabilities"
  ],
  "properties": {
    "agent_id": {
      "type": "string",
      "description": "Stable identifier for this agent profile (UUID, ULID, or similar)."
    },
    "name": {
      "type": "string",
      "description": "Human-readable name of the agent."
    },
    "description": {
      "type": "string",
      "description": "What this agent does and when it should be invoked."
    },
    "model_lineage": {
      "type": "object",
      "description": "Base model and fine-tuning lineage for this agent.",
      "required": ["base_model"],
      "properties": {
        "base_model": {
          "type": "string",
          "description": "Identifier of the underlying model, e.g. 'mistral/Mixtral-8x7B-Instruct'."
        },
        "fine_tuned_from": {
          "type": "string",
          "description": "Optional base checkpoint this agent was fine-tuned from."
        },
        "provider": {
          "type": "string",
          "description": "Provider or platform hosting the underlying model."
        }
      }
    },
    "capabilities": {
      "type": "array",
      "description": "Declared capabilities / skills of the agent.",
      "items": {
        "type": "string"
      }
    },
    "tools": {
      "type": "array",
      "description": "External tools or APIs this agent can call.",
      "items": {
        "type": "object"
      }
    },
    "evals": {
      "type": "array",
      "description": "Evaluation results (benchmarks, test suites, or grades).",
      "items": {
        "type": "object"
      }
    },
    "pricing_model": {
      "type": "string",
      "description": "How this agent is billed (per_call, per_1k_tokens, per_month, etc.)."
    },
    "safety_grade": {
      "type": "string",
      "description": "Optional safety rating or risk classification for this agent."
    },
    "publisher": {
      "type": "object",
      "description": "Entity responsible for this agent (individual or organization)."
    },
    "deployment": {
      "type": "object",
      "description": "Deployment details such as region, uptime, and SLOs."
    }
    // …additional fields omitted for brevity (workplace_tasks, attestations, tags, payment, etc.)
  }
}

This excerpt shows the top-level identity, lineage, capabilities, and commerce/safety surface. The full schema.json on GitHub defines all nested structures and constraints.

What it is

Agent Schema is a JSON Schema–based standard for describing AI agents in a way that is:

Machine-verifiable — suitable for CI checks and registries
Vendor-neutral — not tied to any single platform
Interoperable — designed to map onto existing agent frameworks

It focuses on the “trading card” of an agent — who it is, what it can do, how to call it, and under what constraints — rather than prescribing a specific execution runtime.

How to use it

The canonical schema and examples live in the minuetai/agents repository. Typical flows:

Define an agent profile as a JSON document that conforms to schema.json
Validate it in CI using a JSON Schema validator (e.g. ajv)
Use it as a contract between internal services, registries, and marketplaces

For online validation, you can paste schema.json and an agent JSON into jsonschemavalidator.io.