Summary
Add tracing and observability for LLM calls, so users can monitor, debug and evaluate agentic workflows without relying on print statements or guesswork.
Problem / goal
When running agentic workflows there is no way to observe, trace or debug LLM calls. This makes it impossible to diagnose failures, measure latency, track token usage, or compare prompt versions over time.
Primary user(s)
OpenCode primitive alignment
session.*, mcp (tracing hooks), skills/plugins (evaluation wrappers)
Alignment with VISION/PRINCIPLES/PRODUCT
Observability is a core need for any production-grade agentic tool. Teams cannot responsibly deploy agents without visibility into what they are doing. This directly supports the product's goal of being enterprise-ready.
Testability
pnpm dev + run a multi-step agent task + verify traces appear in LangSmith or OTEL backend (e.g. Jaeger running locally via Docker)
Ready to build it yourself?
Yes
Additional context
Two integration paths to consider:
- LangSmith: direct integration for trace logging, prompt versioning and evaluation dashboards
- OpenTelemetry (OTEL): vendor-neutral open standard routing traces to any compatible backend (Jaeger, Honeycomb, Grafana Tempo etc.)
Even a lightweight OTEL exporter hook would be a great starting point.
Summary
Add tracing and observability for LLM calls, so users can monitor, debug and evaluate agentic workflows without relying on print statements or guesswork.
Problem / goal
When running agentic workflows there is no way to observe, trace or debug LLM calls. This makes it impossible to diagnose failures, measure latency, track token usage, or compare prompt versions over time.
Primary user(s)
OpenCode primitive alignment
session.*, mcp (tracing hooks), skills/plugins (evaluation wrappers)
Alignment with VISION/PRINCIPLES/PRODUCT
Observability is a core need for any production-grade agentic tool. Teams cannot responsibly deploy agents without visibility into what they are doing. This directly supports the product's goal of being enterprise-ready.
Testability
pnpm dev + run a multi-step agent task + verify traces appear in LangSmith or OTEL backend (e.g. Jaeger running locally via Docker)
Ready to build it yourself?
Yes
Additional context
Two integration paths to consider:
Even a lightweight OTEL exporter hook would be a great starting point.