Compare commits

..

13 commits

Author SHA1 Message Date
Anthony Alaribe
b6610cb45c Utils: keep prettyPrintCount's doctest with its definition 2026-11-16 04:25:46 +00:00
Anthony Alaribe
a352b49572 LLM query: float8 casts so TimeFusion returns doubles too 2026-11-16 04:16:22 +00:00
Anthony Alaribe
7a9e095d0c Plan: describe the KPI strip as built 2026-11-16 04:04:24 +00:00
Anthony Alaribe
153cdf8f3f Plan: page routes 2026-11-16 04:04:13 +00:00
Anthony Alaribe
f316e7733c Plan: record deferred follow-ups 2026-11-16 04:03:56 +00:00
Anthony Alaribe
72ed0635dd LLM observability: prefix price lookup, single message decode, shared token/number helpers, group once, capture flag from config 2026-11-16 03:57:42 +00:00
Anthony Alaribe
98651c405f Traces: per-step tokens and cost in the waterfall, LLM cost in the trace header 2026-11-16 03:47:38 +00:00
Anthony Alaribe
05a0f777a3 GenAI ingest: no self-referential op fallback, Claude Code needs prompt.id, only span/log attributes normalised; trace-less records stay separate conversations; capture stays off in .env.example 2026-11-16 03:42:35 +00:00
Anthony Alaribe
c07408c8d1 LLM pages: radiogroup tabs, explicit span and calls labels, full-width title on phones 2026-11-16 03:21:05 +00:00
Anthony Alaribe
2b4c08cb32 Fix GenAI doctests 2026-11-16 03:19:47 +00:00
Anthony Alaribe
a0b45366dd LLM pages: typed finish reasons; single-use renderers live with their caller 2026-11-16 03:18:50 +00:00
Anthony Alaribe
bb50a5d341 LLM pages: one Call record and one query for lists and detail 2026-11-16 03:05:08 +00:00
Anthony Alaribe
ce3514ab5d LLM & agent observability: GenAI normalisation and pricing at ingest, conversations and calls pages, self-instrumented assistant
Every GenAI dialect (OpenTelemetry GenAI, OpenInference, OpenLLMetry, Vercel AI SDK, Langfuse,
Claude Code) is mapped onto the gen_ai.* conventions at ingest, and model calls are priced from
a vendored LiteLLM table. A new LLM & Agents page lists conversations and model calls with
deterministic signals, a transcript and timeline per conversation, Ask AI, and a dashboard.
Monoscope's own assistant now emits invoke_agent, chat and execute_tool spans with real usage.
2026-11-16 02:59:20 +00:00
28 changed files with 1789 additions and 72 deletions