Public beta

Trace every token path your AI agent executes

Token Flame Graph maps each call, tool run, and context rebuild into one session timeline so teams can cut spend and debug faster.

Start free Read integration docs
live session telemetry cost alert active
Tokens
47,832
Cost
$0.86
Loops
3

Why teams use it

Production agent behavior is often opaque. This makes token burn visible and actionable in minutes.

70% of spend can hide in context bloat and retries without clear tracing.
0 alerts in many stacks before loops spiral. We flag them directly in session view.
Slow triage when cost data lacks feature-level attribution. This links spend to workflow.

Three core workflows

Designed for observability, not vanity dashboards.

01 trace

Session flame structure

Inspect every run as stacked spans by turn, tool, and model segments.

02 detect

Loop and waste signals

Highlight repeated execution patterns and context-heavy calls instantly.

03 enforce

Budget controls

Set per-session, per-agent, or per-call limits with warning/block actions.

Integration in minutes

Publish events from your existing PI agent with one collector wrapper.

agent.ts 
import { instrumentAgent, LocalCollector } from "flamegraph-sdk";

const collector = new LocalCollector({
  serverUrl: process.env.FLAMEGRAPH_URL,
  apiKey: process.env.FLAMEGRAPH_API_KEY,
});

instrumentAgent(agent, { collector });

Stop shipping blind cost regressions

Wire session telemetry before your next release and track model spend by real runtime behavior.

Create workspace