Agent Telemetry SDK

Your analytics platform is blind to agents.

Traditional analytics tracks clicks and page views. Nobody tracks what your AI agents actually do — which tools they call, how much they spend, where their reasoning breaks down. Box9 does. Three lines to install. Zero config.

quickstart.py
_

See what your agents actually do.

Every token, every tool call, every reasoning step — measured and visualized in real time.

Events (24h)
0
148 sessions
Total Cost
$0.00
$0.145/session
Avg Latency
0ms
p95: 1,240ms
Error Rate
0.0%
42 errors
Agent Reasoning Funnel
24h
Latency Distribution
p50
Cost by Model
claude-opus-4.6$12.47
claude-sonnet-4.5$4.21
gpt-4.1$3.18
gemini-2.5-pro$1.62
Tool Efficacy Heatmap
6 tools
Events Over Time

Want this for your agents?

Three lines of Python. Full observability in under five minutes.

What Box9 measures.

The Reasoning Funnel

See exactly where agent reasoning breaks down. From initiation through tool calls to completion — every stage is measured.

Token Economics

Know what every agent call costs before your bill arrives. Per-model, per-tool, per-session cost breakdowns.

Latency Distribution

p50, p90, p95 latency across every tool and model. Spot the slow ones. Optimize before your users notice.

Tool Success Rates

Every tool call tracked: did it succeed? How long? How often? Build a reliability profile for your toolkit.

🔗

Agent Receipts

Every action produces a sealed, SHA-256 receipt. Queryable, verifiable, chainable. Agents check before repeating.

Model Comparison

Same task, different models. Compare cost, latency, and success rate. Make data-driven model selection decisions.

SDK Overhead
<2ms
async batching
Lines to Install
3
zero config
Cost Visibility
100%
per-call tracking
Events Tracked
3B+
and counting

The agent era needs new instrumentation.

Box9 is in private beta. Three lines to install. Full observability for every agent call, tool execution, and reasoning step.