Overview

Every agent execution becomes a trace: a complete record of what happened, how long it took, and where it failed. The Lemma SDK emits the trace shape Lemma reads.

One agent execution = one trace. The root is the trace record; LLM calls, tool calls, and app logic are child spans inside it. See the trace contract.

The model

A trace is a tree of spans:

support-agent              <- trace root (input, output)
|- draft-reply             <- generation (model, content)
|- search_docs             <- tool call (args, result)
`- final-answer            <- generation

Traces - the root record for one execution.
Generations - LLM calls.
Tool calls - tool invocations.
Spans - retrieval, ranking, app logic.
Threads & context - conversations, users, metadata.

Pick your path

Instrument with the SDK

Install the Lemma SDK, then add traces, generations, tool calls, and spans.

How Lemma reads your traces

Lemma is an opinionated sink: it reads a specific trace shape to power input/output display, model visibility, timing, tool visibility, threads, and automated issue detection. The trace contract documents exactly what it reads and how each SDK helper maps to it.

Next steps

Setup - install the SDK and point it at Lemma.
Trace contract - the exact shape Lemma reads.
Building high-quality traces - the ideal instrumentation path, with bad → better → best examples.
Troubleshooting - fix missing spans, flat traces, and delayed exports.

⌘I

Getting Started

Instrumentation

Integrations

Connections

The model

Pick your path

Instrument with the SDK

How Lemma reads your traces

Next steps

​The model

​Pick your path

Instrument with the SDK

​How Lemma reads your traces

​Next steps

The model

Pick your path

How Lemma reads your traces

Next steps