brink-sdk 0.0.1

The operations platform for AI agents — SRE primitives for agent reliability

Brink SDK

The operations platform for AI agents — SRE primitives for agent reliability.

Brink brings Site Reliability Engineering to AI agents: circuit breakers that trip on behavioral degradation (not just HTTP errors), SLOs and error budgets for agent-native quality metrics, and chaos testing designed for the failure modes unique to LLM-powered systems.

Installation

pip install brink-sdk

What It Does

Traditional monitoring tells you whether your infrastructure is healthy. Brink tells you whether your agents are behaving correctly — catching hallucination, infinite loops, cost runaway, goal drift, and cascading failures across multi-agent pipelines.

• 🔌 Circuit Breakers — Trip on quality degradation, cost spikes, and step count explosion
• 📊 SLOs & Error Budgets — Quantify reliability with agent-native metrics
• 💥 Chaos Testing — Inject semantic-layer faults to build confidence before production breaks
• 🔭 Structured Observability — Hierarchical tracing (Session → Task → Step → Interaction)
• 🛡️ Budget Guards — Hard limits on cost, tokens, steps, and time per agent run

Quick Start

from brink import monitor

@monitor
def my_agent(query: str) -> str:
    # Your agent code here
    ...

⚠️ Pre-release: This package is under active development. The API is not yet stable.

License

Apache License 2.0 — see LICENSE for details.