AgentBurn getagentburn.com
Sponsor Submit Agent
Home/Live stream

Live agent ingest

Demo live

A schema-accurate replay of what the AgentBurn telemetry SDK streams from a running agent: tool calls, retries, context compaction, task outcomes and the tokens each step costs — at roughly the cadence a production fleet reports.

—
Events / min · rolling 60s
—
Tokens / min · rolling 60s
—
Session tokens
—
Agents active · last 60s
Loading stream scope…
This is a demo stream, not live traffic. Every line is generated in your browser by /assets/js/feed.js?v=1abd15bfdd from the usage profiles in /data/projects.json — nothing is uploaded and no external service is contacted. The event shape (agent, tool call, retry, compaction, task outcome, tokens) is the same one the SDK emits; connect your own agent to replace it with real telemetry.
agentburn / ingest-stream simulated streaming
Starting the stream… Demo data · generated locally in your browser
Right now

Top burners

Full board →

Top five by derived 24-hour tokens (runs per day × tokens per task) for the selected domain.

Connect

Stream your own agent

Read the SDK quickstart

The SDK is a dependency-free ES module. Wrap a task, report tokens and tool calls, and each event arrives in the shape this demo replays — the same fields the boards on this site are computed from. Works with LangGraph, CrewAI, LlamaIndex or a hand-rolled loop.

  • Tokens, tool calls, retries and the task outcome — the four numbers behind every board.
  • Anonymised by default: no prompts, no file contents and no customer data leave your process.
  • Cost is priced from the public model price table, so you never have to report dollars yourself.
javascript 
import { createClient } from '/sdk/js/agentburn.js';

const burn = createClient({ project: 'your-agent' });

await burn.trackTask(runAgent(), {
  model: 'claude-sonnet-5-5',
  task: 'fix-flaky-test'
});