Loading…
GMAN Live Demo
Real telemetry from an autonomous AI agent.
Real telemetry from an autonomous AI agent.
Live tracing, cost attribution, and decision telemetry for autonomous LLM workflows. This page instruments one of my own production agents (social-agent), traced end-to-end by my own SDK.
// LLM systems are stateful, probabilistic, and increasingly autonomous. // Application logs cannot show reasoning flow, cost propagation, retry // semantics, or tool interactions across agentic workflows. GMAN does.
The agent reviewed 180 candidate commits. 75% scored below threshold and auto-deflected; the human only saw the 25% that cleared the gate.
$0.33 per generated tweet · $0.08 per session average
| path | p50 (ms) | p95 (ms) | p99 (ms) | mean (ms) |
|---|---|---|---|---|
| direct | 0.118 | 0.183 | 0.273 | 0.13 |
| proxied | 0.334 | 0.548 | 1.086 | 0.36 |
| delta · proxy tax | +0.216 | +0.365 | +0.813 | +0.23 |
deployment 1 Railway replica · free Supabase · 100 req/min ceiling · no queue · us-west
97/min sustained · drop 7.14% · 357 × 429 · p50 86.34ms · p95 248.66ms · p99 387.07ms
Constraints listed inline are the deployment envelope; numbers are valid inside it. Full methodology →
score 1 < 5 threshold; no draft generated · pipeline cost $0.0187
There is no new activity or story to share since there have been no recent commits.
score 0 < 5 threshold; no draft generated · pipeline cost $0.0187
There is no new activity or story to share since there are no recent commits.