Skip to main content
Lemma turns each agent execution into a trace, finds the executions that went wrong, and groups the repeat offenders into issues you can triage. This page walks through that path end to end: how a trace becomes an issue, how projects and agents relate to each other, and which dashboard surface to open at each step. For the SDK vocabulary (trace, span, generation, tool call, thread, project), see SDK concepts. For the dashboard vocabulary (issues, occurrences, artifacts, agents), see Platform concepts. To send your first trace, see Quickstart.
Project dashboard showing to-do items, headline stats, and resolved-versus-dismissed and traces-per-month charts

​
How a trace becomes an issue

An instrumented agent emits one trace per execution, or per conversation turn when you set a thread ID. See Concepts for how Lemma turns that trace into an issue — the detection timing, the probabilistic grouping, and the vocabulary for both. Once an issue exists, you investigate it in the dashboard or through Inspect, then resolve, dismiss, or route it to Slack, Linear, MCP, or webhooks.

​
Projects and agents

You create projects to isolate traces by environment (dev, staging, prod) or product line; each has its own dashboard, API key, and settings. Agents aren’t registered the same way — see Concepts for how a trace’s agent name scopes Artifacts, Analytics, and issue grouping.

​
The dashboard surfaces

Each surface below answers a different question about the same underlying traces: A typical investigation starts on Analytics or Issues, where you spot a trend. From there you open the supporting traces, ask Inspect what’s going on, add Artifacts context if detection missed the mark, and alert or ticket the finding through Connections.

​
What to do after the first trace

  1. Open Traces in the project and confirm the execution tree and conversation look complete.
  2. Wait for issues if the conversation is still active. Extraction runs after a thread goes quiet, or immediately when there’s no thread ID.
  3. Add Artifacts context for that agent if domain rules matter to how you’d judge a run.
  4. Connect Slack, Linear, the MCP server, or webhooks once the team needs those paths. See Connect your workflow.

Concepts

Traces, issues, occurrences, artifacts, and agents.

Analytics

Project-level volume, latency, errors, tools, models, and cost.

Inspect

Ask questions about the trace, issue, or page you’re on.

Traces

Search and inspect ready executions in the dashboard.

Issues

Recurring failure patterns with status and evidence.

Artifacts

Per-agent context, understanding document, and decision-flow diagram.

Connect your workflow

Slack, Linear, MCP Server, and webhooks.