Distributed traces

Waterfall every run, down to the token.

Each trace is one top-level call, expanded into a full span tree — nested agents, tools, and LLM calls, with the exact prompt and response for every step.

support-triage5.84s · 8 spans
support-triage
classify
fetch-order
search-kb
rerank
draft-reply
The full tree

Nested spans, not flat logs.

Foglamp reconstructs parent/child relationships across agents, tools, and models — so a multi-hop run reads as one coherent waterfall instead of a pile of disconnected log lines.

  • Agent → tool → LLM nesting preserved
  • Per-span duration, tokens, and cost
  • Errors highlighted exactly where they happened
support-triage
classify
fetch-order
search-kb
rerank
draft-reply
Replay

See the exact prompt and streamed response.

Open any span to read the full system prompt, messages, and the response as it streamed — including time-to-first-token. No more reproducing bugs blind.

  • Full prompt + response payload per span
  • TTFT and token-by-token streaming replay
  • Copy any message straight into a repro
Span timing
520ms
TTFT p95
2.46s
draft-reply
4.2k
tokens

Your agents are running in the fog.

Cost, latency, errors, eval scores — all there, all invisible. Wrap your model and turn the light on.

Start free