ReAct Agent

A tool-using agent with labels, budget constraints, and a configured action space.

schema_version: "1.0.0"

metadata:
  id: "research-assistant"
  name: "Research Assistant"
  version: "1.0.0"
  description: "An agent that researches topics using web search"
  labels:
    team: "ai-platform"
    environment: "development"

interface:
  input:
    type: object
    properties:
      query:
        type: string
        description: "The user's research question"
    required: [query]
  output:
    type: object
    properties:
      response:
        type: string
        description: "The researched answer"
    required: [response]

execution_policy:
  id: agf.react
  config:
    instructions: |
      You are a research assistant. When the user asks a question:
      1. Use the web_search tool to find relevant information.
      2. Synthesize the results into a clear, cited answer.
      Be specific about the agent's role, capabilities, and constraints.
    model: "gemini-2.0-flash"
    provider: "google"
    max_steps: 10
    tool_choice: auto

action_space:
  local_tools:
    - alias: web_search
      description: "Search the web for information on a topic"

constraints:
  budget:
    max_token_usage: 100000
    max_duration_seconds: 300

What's New Here

Labels

labels:
  team: "ai-platform"
  environment: "development"

Key-value pairs for filtering, routing, and organizational tagging. Useful for multi-team deployments where you need to query agents by owner or environment.

ReAct Config Options

Field	Description
instructions	System prompt (multi-line with |)
model	Model identifier
provider	Model provider hint (optional)
max_steps	Maximum reasoning steps before the agent must return
tool_choice	auto, required, or none

Action Space

The action_space section declares what the agent can do. Here we define a single local tool:

action_space:
  local_tools:
    - alias: web_search
      description: "Search the web for information on a topic"

The alias is the name the LLM sees. The runtime provides the actual implementation. This separation is key to portability — the same definition works on any runtime that implements a web_search tool.

Budget Constraints

constraints:
  budget:
    max_token_usage: 100000
    max_duration_seconds: 300

Hard limits enforced by the runtime:

• max_token_usage — total token budget across all LLM calls
• max_duration_seconds — wall-clock timeout

If either limit is exceeded, the runtime terminates the agent gracefully.

Validate

info

The agf CLI will be publicly available soon. See the CLI Reference for the command reference. You can validate this example now using the Playground.

agf lint research-assistant.agf.yaml
# ✓ research-assistant.agf.yaml: 0 errors, 0 warnings