Skip to main content
These pages are for anyone who sends agent runs to Atlan Trident and reads them back. A trace is the step-by-step record of one agent run: its model and tool calls, tokens, cost, and errors. Send traces from your agent, then find each run under its owner, workspace, and session. The API key that sends a trace decides whose it is, and the workspace header decides where it lands; see How trace attribution works. A session groups a run, or a conversation’s runs, under its agent, and a visitor is the person or account a run was for. Highlights and Full trace present each run; see How trace inspection works.

Core offerings

Send and verify traces

Send as the right identity and confirm arrival.

Inspect recorded run

Read highlights, then follow the full trace.

Group runs into sessions

Tie a conversation’s runs together.

Identify visitors

Record who each run was for.

Compare agent usage

Compare spend, runs, and errors across agents.

Choose where to start

Traces can carry prompts, tool arguments, and responses. Decide what leaves your process with Control trace privacy, and keep customer data out of test traces. Atlan Trident never moves a trace after ingest: a run sent with the wrong key or workspace stays where it landed. Ingest has size and span-count limits per request; see Trace limits and privacy. The Claude Code trace plugin is in preview; see Capture coding-agent sessions.

See also

  • Trace attributes: Which attributes set agent, session, visitor, skill, model, tokens, and cost.
  • Configuration: Every SDK tracing option and environment variable.
  • Public API: POST /otel/v1/traces and the agent, session, and user trace reads.