Inference Gateway

Appearance

Agent Definition Language (ADL) ​

Agent Definition Language (ADL) is a declarative YAML format for defining an A2A (Agent-to-Agent) agent as code. A single agent.yaml manifest captures everything about an agent - its provider and model, the tools it can call, the skills it knows, its server and authentication settings, and its CI/CD and deployment wiring - so the whole agent is version-controlled, reviewable, and reproducible instead of scattered across hand-written boilerplate.

ADL is the declarative layer that sits above the Agent Development Kit (ADK). You describe the agent once in ADL, and the ADL CLI generates a complete, enterprise-ready project - in Go or Rust (TypeScript planned) - that builds on the ADK runtime. Regenerating after an edit preserves your custom code through .adl-ignore, so the manifest stays the single source of truth as the agent evolves.

Canonical reference. The full ADL schema, field reference, and authoring guide live at adl.inference-gateway.com. This page is a conceptual overview; the ADL CLI reference documents the tooling that consumes ADL files.

What an ADL file looks like ​

Every ADL manifest is a YAML document that starts with an apiVersion and kind, then describes the agent under spec:

yaml
apiVersion: adl.inference-gateway.com/v1
kind: Agent
metadata:
  name: my-weather-agent
  description: 'Provides weather information'
  version: '0.1.0'
spec:
  capabilities:
    streaming: true
  agent:
    provider: deepseek
    model: deepseek-v4-flash
    systemPrompt: 'You are a helpful weather assistant.'
  tools:
    - id: get_weather
      name: get_weather
      description: 'Get current weather for a city'
      tags:
        - weather
      schema:
        type: object
        properties:
          city:
            type: string
            description: 'City name'
        required:
          - city
  server:
    port: 8080
  language:
    go:
      module: 'github.com/example/my-weather-agent'
      version: '1.26.2'

Scaffold, validate, and generate a runnable project from a manifest like this with the ADL CLI:

bash
adl init my-weather-agent      # interactive wizard writes agent.yaml
adl validate agent.yaml        # check it against the schema
adl generate --file agent.yaml --output ./my-weather-agent

See the ADL CLI reference for the complete command surface and a full field-by-field schema breakdown.

How ADL fits the ecosystem ​

  • ADL - the declarative YAML contract that describes an agent. Canonical spec: adl.inference-gateway.com.
  • ADL CLI - reads an ADL file and generates a runnable A2A agent project, including tool stubs, skills, CI/CD, and deployment manifests.
  • Agent Development Kit (ADK) - the Go, TypeScript, and Rust runtime libraries that generated projects build on.
  • A2A - the Agent-to-Agent protocol the generated agent speaks, so the gateway and its CLI can discover and call it.