Every agent run,
fully visible.

Every failure caught, every dollar traced. AI tells you exactly what to fix before costs compound.

How it works

Open source SDK

Two-line install

Works with any framework

Your agents are running. You have no idea how they are performing.

The first sign is usually a surprise invoice.

$2,400burned before you noticed it.

1 in 3agent failures go undetected until a user reports it.

0visibility into which call caused it.

Complete visibility
into every agent run.

From the first run. No configuration. No sampling.

PerformanceHow long every step takes and exactly where your agent slows down.

CostWhat each run costs, which model is spending the most, and why.

QualityWhich runs failed, what the error was, and how often it happens.

ReliabilityWhen retry storms hit, how bad they got, and what triggered them.

BusinessWhether your agents are worth what you're paying to run them.

From install to insight in minutes.

0101

Install the SDK

Two lines. Works with LangChain, CrewAI, LlamaIndex, or any custom agent.

from agentmetrics import sentinel

sentinel.configure(api_key="am_...")

@sentinel.track(agent_id="research-agent")
def research(query):
    return llm.complete(query)

0202

Live from run one

Every token, tool call, retry, and latency appears on your dashboard. Nothing to configure.

Tokens4,218 in / 891 out

Latency2.3s avg

Retries12 detected

Cost$0.031 / run

0303

Fix it. Watch costs drop.

AI tells you exactly what to change and why. One fix at a time.

Switch to claude-3-haiku for 90% of runs

Save $612/mo

Before

$1,240/mo

After

$498/mo

Free SDK. Managed cloud when you need it.

Upgrade when your agent spend justifies it.

Free SDK

Freeforever

Self-hosted. Always free, no limits on self-hosted usage.

1 tracked agent

10,000 events/month

7-day history

Open source dashboard

Every agent run, fully visible.