Agent
An agent is an AI worker in your organization — Claude Code running for a developer, a Cursor instance, an autonomous agent, or Claude Desktop. Each agent has an owner (a person), a model, and a history of sessions and decisions. Over time an agent accrues a profile: its volume, spend, and real outcomes. That profile is how you decide which agents to scale, optimize, or retire. See Agent profiles.Session
A session is a single run of an agent — one coherent stretch of work with a unique ID. A session breaks down into its decisions and carries aggregate stats: decision count, cost, duration, tool usage, and quality scores. Sessions also get a health view with metric trends, outliers, and drift indicators.Decision
A decision is one reconstructed model call — the atomic unit of the decision ledger. Each decision has:- a title and outcome (Outcomes & Intent),
- a logic trace of reasoning steps,
- evidence backing each step (Evidence & the Logic Trace),
- quality metrics, cost, and duration.
Agent → session → decision on the Explore canvas