Unifying app spans, Sentry errors and LLM calls into one trace for AI debugging
dank_as_fuck_ · reddit · 2026-09-27
The author found most AI bugs don't live in the model call: exceptions sit in app code, bad inputs come from tools, and the model call in between looks fine — forcing manual timestamp-stitching across Sentry, OTel backends, and LLM tools. He built axonpush.xyz: keep your existing OTel exporter or Sentry DSN, and app spans, DB/queue spans, exceptions, and model/tool calls land on one trace, with exceptions grouped into issues linked to their traces. Failed runs can be handed to Claude Code or Codex via MCP with the trace as context. Hard parts: aligning Sentry events and OTel spans on what one 'request' is, and deduplicating error groups.
More from coding & agent
- omarsar0 on custom agent benchmarks: public numbers don't mean much, Jev Router rivals GPT-6 Astra low — omarsar0 · 2026-09-27
- Opus 5.5 + Meta Ads MCP builds a full ad campaign in 47 minutes, claims 70% cost cut — PrajwalTomar_ · 2026-09-27
- Agent chat UI design debate: the timeline is strictly chronological, cards reflect current state — zeeg · 2026-09-27
- Microsoft's ProgramDistill turns interactive web apps into verifiable SWE training tasks — _akhaliq · 2026-09-27
- Benchmarking 'Decision Models': Scoring Options Is 7-54x Faster Than Generating Probabilities — vykthur · 2026-09-27
- Deploying AI Agents? Start With One Workflow, Warns Architect — DavidLinthicum · 2026-09-27