r3fresh

r3fresh ALM SDK

Agent Lifecycle Management SDK – A Python SDK for tracking AI agent execution with policy enforcement, event emission, and structured event data for downstream analytics.

Overview

The SDK provides automatic instrumentation for AI agents, capturing:

  • Tool calls with policy enforcement (allow/deny) and latency tracking
  • Run lifecycle (run.start / run.end) with summary statistics
  • Tasks (task.start / task.end) for logical units of work
  • Handoffs for agent-to-agent transitions
  • Structured errors (type, message, source, retryable) in tool and run events
  • Version tracking (schema, SDK, agent, policy) on every event

All events are emitted automatically. They can be sent to stdout (development) or an HTTP endpoint (production). The SDK does not perform analytics itself; it produces events for your backend or analytics pipeline.

Quick Start

pip install r3fresh
from r3fresh import ALM

# Initialize the SDK
alm = ALM(
    agent_id="my-agent",
    env="development",
    mode="stdout",  # or "http" with endpoint
    agent_version="1.0.0",
)

# Define tools with automatic policy enforcement
@alm.tool("search_web")
def search_web(query: str) -> str:
    """Search the web for information."""
    return f"Results for: {query}"

# Run your agent with automatic tracking
with alm.run(purpose="Process user query"):
    result = search_web("Python SDK documentation")
    print(result)

# All events are automatically captured and emitted

Full Quick Start guide →

What's Next