1. Docs
2. /
3. Tool reference
4. /
5. Agents & Machines

Agents & Machines

Dispatch agents to your own attached hardware, watch runs, and collect results.

31 verbs in this family.

• drainMachine write Drain (or un-drain) one of your machines: the safe way to take a box offline without losing agent work.
• fleetDrainStatus read Read the drain state of your machines plus every agent run that stopped short and can be picked up.
• getAgentResult read Read the outcome of a dispatched agent session by its conversationId: the orchestrator readback that completes the dispatch→readback→synthesize loop.
• gradeRun write Grade one of your agent runs: record whether it was "useful", "partial", or "failed", plus an optional failure mode when it fell short ("grounding" | "instruction-following" | "format" | "tool-use"…
• handoffSession write Hand this session off to a successor: write the baton, open the next session on the SAME machine and seat, and mark this one handed off: one verb instead of the ritual.
• listAgentSessions read List your agent sessions on the live roster: every agent currently queued, running or PARKED across your machines, PLUS any lane DROPPED or LOST (its machine restarted mid-run) in the last 24 hours:…
• listMachines read List the machines attached to your account that can run agents: each with its name, lifecycle status, derived liveness, last-seen heartbeat, browser readiness, and latest telemetry.
• listSessions read List your sessions as the SESSION BUS sees them: the roster you address `messageSession` from.
• messageSession write Send a message from one of YOUR sessions to another one, across machines.
• noteSession write Write (or clear) YOUR own note on one of your agent sessions: one free-text line about what the run was for, what it got wrong, or what to do with it next.
• readAgentOutput read Read back what a dispatched agent session actually PRODUCED: its full transcript, final output, and tool-call log: by conversationId.
• readDeliveryFlow read Read how long each of your fleet's deliveries actually took to move, one row per delivery (a ledger Task and every lane that worked it), with every value linked back to the receipt it came from.
• readFleetRegeneration read Read the last self-regeneration reports for your fleet, newest first, without running a new pass.
• readSeatScoreboard read Score every model on every seat on every attached machine over a window, off the runs themselves: how many lanes each one was given, how those lanes ended (completed, versus stopped, dropped,…
• readSessionSweeps read Read the last sessions sweeps, newest first, without running a new one.
• readStartupReceipt read Read your own startup receipt: the bounded, dated proof of what this seat can and cannot do right now, before it leads anything.
• recheckSeatSignIn write Ask one of your machines to RE-PROBE its agent seats right now: whether the Claude, Codex and Cursor CLIs are installed, signed in, and able to complete a one-turn smoke.
• regenerateFleet write Run a self-regeneration pass over the fleet RIGHT NOW: read every attached machine and every seat on it, classify each seat into exactly one class, carry out the repairs that are safe to make…
• reportPresence write Names your own presence node in the user's Agents panel.
• resumeAgentRun writeexternal Pick up an agent run that stopped short: respawn it on the same machine, continuing the SAME branch, seeded with a preamble pointing at the work its predecessor left behind (so it continues rather…
• runAgent writeexternal Start (spawn) an autonomous agent on one of your machines to carry out a task.
• runOnMachine writedestructiveexternal Run a single shell command on one of your machines (a MachineSync box) through The Cloud, and stream its output back into the Agents view: box ops without a physical terminal (update the daemon,…
• setDispatchSeatPolicy write Set the account's STANDING DISPATCH SEAT LAW: which seat a spawn takes when the caller names no model.
• setLaneCap write Set how many agent lanes one of your machines runs AT ONCE: the pace knob for a box that is running hot.
• setMemoryGuard write Read or adjust the memory guard on one of your machines: the rule that stops a box CLAIMING new agents while it has no memory to give.
• setProjectStatsVisibility writeconsent Request one project's live Agents view public, or revoke it immediately.
• setSeatMode write Set HOW one of your agent seats runs once a lane lands on it: the fast-or-thinking dial, the model tier, and the reasoning effort.
• steerAgent write Amend a QUEUED or RUNNING agent session in flight: add an instruction to a lane instead of stopping it and firing a fresh one.
• stopAgent write Stop one of your running agent sessions: a graceful halt of an agent you started, for a runaway, duplicate, or no-longer-needed run.
• sweepSessions write Reconcile every session that ended badly against what it ACTUALLY produced, and repair what can be repaired.
• tagSession write Add tags to or remove tags from one of YOUR agent sessions: the exact mirror of `tagPage`, applied to a dispatched run instead of a page.