Debug local LLM and agent traces with Axon
Run Axon as a local OpenTelemetry endpoint and dashboard for inspecting LangChain and instrumented agent traces without sending run data to a cloud service.
npx skills add agentskillexchange/skills --skill debug-local-llm-and-agent-traces-with-axon
npm install -g @axon-ai/cli, run axon-ai init --project inside the target project, then run axon-ai start. Point LangChain, OpenLLMetry, OpenInference, or raw OTEL exporters at http://localhost:4000/v1/traces or the configured port, then inspect traces in the local dashboard.Use Axon when an operator needs a local trace collector and dashboard for LLM or agent runs during development and debugging. The workflow is to install the Axon CLI, initialize a project, start the local backend and dashboard, point LangChain, OpenLLMetry, OpenInference, or raw OpenTelemetry exporters at the local OTLP endpoint, then inspect transcript, tree, waterfall, raw span, token, and cost views while the agent runs. Invoke this instead of using an observability product normally when the task is short-loop local debugging, trace inspection, or privacy-preserving development capture for an instrumented agent workflow. The boundary is local OpenTelemetry trace ingestion, SQLite-backed storage, and dashboard analysis for development agent runs; it is not a generic LangChain SDK, hosted monitoring platform, or full production evaluation system.
What this skill actually does
Inputs and prerequisites: Node.js, `@axon-ai/cli`, a LangChain or OpenTelemetry-instrumented application, and permission to create the local `.axon-ai` project directory and trace database.
Setup notes: Install `@axon-ai/cli`, run `axon-ai init –project ` in the project directory, start the local service with `axon-ai start`, and configure the application OTLP exporter for `http://localhost:4000/v1/traces` or the selected port. Use `axon-ai status` and `axon-ai stop` to manage the local service.
Source and verification boundary: use https://github.com/langchain-tracer/Axon and https://www.npmjs.com/package/@axon-ai/cli as the canonical references before running the workflow; keep instrumentation, endpoint configuration, and trace database handling reviewable against upstream docs.
Framework fit: publish this as a Multi-Framework workflow because Axon accepts standard OpenTelemetry/OpenLLMetry traces from multiple agent stacks. Do not map it to MCP: MCP is not central to upstream invocation.