Agent Ops
Agent Ops is the session runtime: the surfaces where agent work happens live. A session — called a thread in the UI — opens here, runs under an environment policy, pauses for approvals when actions carry risk, and closes with a recorded outcome. Everything in this section describes that happening work — not the configuration behind it.
Open in product: preshos.com/agent-ops
Agent Ops is the session runtime: the surfaces where agent work happens live. A session — called a thread in the UI — opens here, runs under an environment policy, pauses for approvals when actions carry risk, and closes with a recorded outcome. Everything in this section describes that happening work — not the configuration behind it.
The overview lists open sessions with goal, environment, and status per row. Each row opens the session it names: transcript, preview, and reply box for that run. Sessions close with a recorded outcome; the list keeps what is running and what already closed.
- One goal per session; a second task means a second session.
- Scope comes from the environment and agent picked at the start.
- Risky actions pause at the approval gate inside the session.
The runtime in order
Section titled “The runtime in order”- Starting a session: agent and environment selection, and what each choice changes.
- Sessions: what a session is once work starts.
- Threads: what a thread is inside a session.
- Model selector: which model the session reasons with.
- Thinking: the visible chain of thought behind agent actions.
- Sandbox: the isolated execution space sessions run inside.
- Working folder vs artifacts: scratch space against reviewable outputs.
- Artifact versions: how outputs iterate without losing the record.
- Computer use: OS-level control from inside a session.
- Browser use: scoped browser work from inside a session.
- Plugins in session: how tools and skills get invoked while work runs.
- RBAC in Agent Ops: who may use which agent, run which actions, and open which sessions.
- Superuser: the elevated mode for work that needs it.
What lives elsewhere
Section titled “What lives elsewhere”- Agents: the configured inventory behind sessions — agents, environments, models, tools, skills, plugins, MCP, memory, providers, and knowledge.
- Automations: scheduled and connector-driven work that fires without a live session.
- Approvals: the human gate — queue, artifact review, and the allow/deny record.
- Settings: tenant identity, membership, and security around all of the above.
The short mental model
Section titled “The short mental model”The console scopes the work (environment, agent, work item). The session does the work under the environment policy. The gate pauses risky actions for a human decision. The outcome keeps the record. One goal per session; a second task means a second session.
New to sessions? Start with Your first session, then run day-to-day with Working with sessions.