sentinelai-sdk 0.1.0

Drop-in reliability observability for multi-agent AI workflows

sentinel-ai

Drop-in reliability observability for multi-agent AI workflows.

Install

pip install sentinel-ai

Quickstart

import sentinel

sentinel.init(api_key="sk_live_...")

# Decorator — add to any existing function
@sentinel.trace_step(name="research-agent", step_type="llm_call", workflow_name="My Pipeline")
def research_agent(query: str) -> dict:
    return my_llm.run(query)  # unchanged

# Auto-patch OpenAI client
sentinel.patch_openai(client, workflow_name="My Pipeline")

# Auto-patch Anthropic client
sentinel.patch_anthropic(client, workflow_name="My Pipeline")

# LangChain callback
cb = sentinel.LangChainCallback(workflow_name="My Pipeline")
llm = ChatOpenAI(model="gpt-4o", callbacks=[cb])

Get your API key at www.agentsentinelai.com/dashboard → ⚙️ Settings.

Integration patterns

Pattern	Lines of code	Best for
@sentinel.trace_step	1 line per function	Standalone agent functions
sentinel.patch_openai(client)	2 lines total	Teams using OpenAI SDK
sentinel.patch_anthropic(client)	2 lines total	Teams using Anthropic/Claude SDK
sentinel.LangChainCallback()	1 line total	LangChain / LangGraph users
sentinel.workflow() context manager	5 lines total	Any existing pipeline

What you see in the dashboard

Tab	What you see
Workflows	Execution DAG: Agent A → B → C
Traces	Per-step latency, input/output payloads
Incidents	Auto-created when any step raises an exception
State	Shared state written by agents
Contracts	Validated handoffs between agents