→ 01 · ZERO LATENCY OVERHEAD
Async logging. TTFT stays clean.
Tracing writes happen asynchronously so instrumentation does not slow Time to First Token on your live agent paths.
─── AGENT TRACE · WAITLISTCOMING SOON
TokenGO
Agent trace.
Full agent execution trees, root-cause debugging, and per-step cost, included with TokenGO inference, with zero TTFT overhead.
- Async logging
- OTel native
- Included
─── 02 · THE HOOKLATENCY · COST · STANDARDS
Tracing that doesn't tax your agents.
→ 02 · NO EXTRA SUBSCRIPTIONS
Included with your LLM compute.
Stop paying hundreds a month for separate tracing software. Agent tracing ships natively with TokenGO inference, free while TokenGO is your provider.
→ 03 · OPENTELEMETRY NATIVE
Open standards, not a lock-in SDK.
Full compatibility with open standards and OpenTelemetry SDKs so traces plug into the tooling your team already knows.
─── 03 · KEY FEATURESTREES · ROOT CAUSE · METRICS
What you open in the dashboard.
One console for the full agent run: expandable trees, the step that broke, and exact token and latency numbers on every hop.
Trace · support-agent-run
Failed · step 4 · 2.4s
Span tree
- TRACEagent.run2.41s
- SPANretrieval.search340ms
- TOOLtool.lookup_order210ms
- FAILgeneration.reply1.12s
- MODELmodel.chat1.05s
Timeline
0ms1.2s2.4s
Generationgpt-class · 842 tok
Input
User asks about order #4182 delay. Context: warehouse status + prior reply.
Root cause
Tool returned an empty order payload; model hallucinated a tracking link on step 4.
Scores
Helpfulness0.41Grounded0.12Latency OKpass
Sessions · users · metadata tags · cost per run
→ 01 · TRACE TREES
Full execution graphs.
Expandable step-by-step visual trees of complex agent loops, tool invocations, and sub-agent handoffs.
→ 02 · ROOT CAUSE
Pinpoint the break.
Instantly see which prompt, tool output, or hallucination caused an agent to fail on step 12, not after hours of log archaeology.
→ 03 · COST & LATENCY
Per-step metrics.
Exact token usage and millisecond-level breakdown per step and per model, on the same timeline as the run.
─── 04 · HOW IT WORKSSWAP · RUN · DEBUG
Three steps to live traces.
- Step 01
Change one line.
Swap your base URL or import the lightweight TokenGO SDK.
- Step 02
Run your agent.
Execute your normal agent loops and tool calls. No special sampling mode required.
- Step 03
Debug live.
Open your dashboard to view full, unsampled trace graphs in real time.
Rolling out in batches. Claim a key for priority access.