Guides
Each guide does one job from start to finish, with code you can run against a local daemon. Start with the guide for the framework your agent uses, then pick the job you need next.
Last updated
Live trace 8 spans 3,430 tok 3.18 s $0.0098
Source Span
01s2s3s
- TS Vercel AI SDK POST /api/chat Vercel AI SDK · TS
- TS Vercel AI SDK gpt-4.1-mini Vercel AI SDK · TS
- TS Vercel AI SDK search_orders Vercel AI SDK · TS
- Go OpenTelemetry GET /orders/{id} OpenTelemetry · Go
- Py LlamaIndex vector.query LlamaIndex · Py
- Py LangChain refund_agent LangChain · Py
- Py Anthropic SDK claude-sonnet-4 Anthropic SDK · Py
- Py LangChain issue_refund LangChain · Py
- How to log OpenAI API calls with tokens and cost Every chat completion with its prompt, answer, tokens and cost, streamed or not.
- How to trace LangChain and LangGraph agents OpenInference and one OTLP exporter: every chain, model call and tool in the waterfall.
- How to trace the Vercel AI SDK AI SDK telemetry over OTLP, with prompts, tool calls and cost totals that add up.
- How to trace Pydantic AI agents with OpenTelemetry Agent.instrument_all() and an OTLP exporter, with prompts you can search.
- How to group LLM traces into conversations One trace per turn, one session id per conversation, and the queries that read it back.
- How to track LLM cost per user Tag runs with the end user, then break spend down by user without storing who they are.
- How to collect thumbs-up and thumbs-down feedback on LLM answers A vote from your product lands on the exact run that produced the answer.
- How to send LLM alerts to Slack A monitor on error rate or spend, and a 30-line relay that posts it to a channel.
- How to run LLM regression tests in GitHub Actions Score a dataset on every pull request and fail the job on a real regression.
- How to add a human approval step to an AI agent A durable wait that holds the run until a person says yes or no.
- How to read agent traces in Claude Code and Cursor Give a coding agent a read key and three tools: find a run, open it, compare it.
For the concepts behind them, read the Concepts page. Every endpoint the guides call is in the HTTP API reference.