Display TensorRT-LLM AgentX server metrics

[cquil11]

Summary

• add a TensorRT-LLM server-metrics adapter for Dynamo disaggregated roles
• expose TensorRT-LLM token, cache, queue, and KV metrics in AgentX charts and aggregates
• preserve separate prefill and decode metric sources

Follows SemiAnalysisAI/InferenceX#2625 and NVIDIA/TensorRT-LLM#12545.

Validation

• 24 focused Vitest tests passed
• oxlint passed
• oxfmt check passed
• TypeScript typecheck passed

---

Note

Medium Risk
Changes versioned ETL and aggregate extraction paths used by the agentic detail API; incorrect mapping could mislabel disagg workers until backfill, but scope is additive framework support with focused tests and no auth or data-mutation logic beyond JSONB self-heal.

Overview
Adds TensorRT-LLM (trtllm_*) alongside existing vLLM/SGLang fallbacks so AgentX chart series and aggregate stats populate KV utilization, prefix-cache hit rate, queue depth, prefill/decode TPS, host KV usage, and prompt-token source splits from native TRT-LLM scrapes.

Introduces a trtllm server-metrics adapter (selected when the framework string contains trt) that maps disaggregation_mode / dynamo_component to canonical prefill vs decode roles. For disaggregated runs, per-endpoint metricSources are emitted with endpoint-level role hints when labels are incomplete.

Bumps CHART_SERIES_VERSION to 13 and STATS_VERSION to 8 so backfill recomputes stale precomputed JSONB; oversized blob stream-parse key sets include the new TRT-LLM metric names.

^{Reviewed by Cursor Bugbot for commit 83d2f0b. Bugbot is set up for automated code reviews on this repo. Configure here.}

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
inferencemax-app	Ready	Preview	Aug 17, 2026 4:19pm