Skip to content

Threads

Open in product: preshos.com/agent-ops/threads

A thread is the message-and-action line carrying one goal: the prompts, agent steps, tool calls, drafts, and decisions in the order they happened. Thread is the UI name — storage and the API call the same run of work a session. Either name points at the same thing: one bounded run under one environment and agent scope.

  • Thread is what members see and open. Session is what storage and the API record. A thread reread tomorrow is the session record — nothing is lost between the two names.
  • Run is one attempt inside the thread: the agent tries, the transcript keeps every try, the outcome records the result.
  • The harness identity behind the run (Eve session) is platform-side. Members operate the thread; the harness name only matters when reading platform logs with support.
  • A thread starts with a bounded goal and ends with a recorded outcome. The goal does not change mid-thread; a new goal means a new thread.
  • Reused threads drift. Continuing yesterday’s thread for today’s task drags old context into new decisions — start fresh instead.
  • The thread transcript is the audit line: what was tried, what was approved or denied, what changed. It reads the same for the member, the admin, and the record.
  • The thread inherits its environment: runner, repository linkage, sandbox timeout, tool grants, and policy come from the session scope, not from the wording of any single message. The repository comes from the environment, not the agent.
  • A thread that cannot reach a system is a scoping problem, not a prompt problem. Check the environment, not the phrasing.
  • Threads belong to their tenant and org scope. Session records are not portable across tenants.