Skip to main content
A loop is what makes an agent act on its own. It is a playbook of what the agent does on each tick, run at a frequency you set, and it lives in the agent’s folder at loops/<loop>/loop.md. An agent can own several loops. An agent with no loop of its own gets a generic Default loop, driven by its AGENT.md, the first time you start it. Each time you start a loop in Loop mode, Condor opens a new session: a numbered run with its own journal, its own per-tick snapshots and its own executors. This page shows you how to watch a session, pause it, stop it and read what the agent did on each tick. For what a session is and what Condor writes to disk, see Sessions.

The agent page

Open an agent from Agents in the Workspace, or go to /agents/<slug>. Choose a loop in the selector under the agent’s name.
The Market Making Expert's agent page with the PMM Mister Operator loop selected and the Now, Runs, Fleet and Playbook tabs
The page has four tabs:

Start a session

Click Start to open the Start New Session dialog.
The Start New Session dialog with Loop mode selected
Execution Mode decides what the run does: Dry Run and Run Once don’t create a session. Each writes one result that you read in the Runs tab.
After a Dry Run or Run Once, the Now tab still says “This loop has not run yet”. That is expected: Now follows loop sessions. Open Runs to see the result.
The agent page after a dry run: the run selector shows D2 with 1 tick and the Now tab still says the loop has not run yet
The other fields:

Pause, resume and stop

While a session runs, the Start button is replaced by Pause and Stop.
Stopping a session doesn’t close its positions. When you start the same loop again, the new session adopts the executors and bots the old one left running. Held positions and Run Once executors are not adopted. See Stopping a loop.
To stop the loop and wind its positions down, ask the agent in chat to shut its loop down. That runs an emergency shutdown, which closes this session’s executors according to the agent’s shutdown policy. To change whether a loop starts again after Condor restarts, use the resumes on restart / stops on restart toggle in the Playbook tab. It applies to the running session and to future ones.

Read what the agent did

Open the Runs tab and pick a run. Each tick shows:
  • System Prompt: everything the agent was given for that tick
  • Agent Response: its reasoning and decision
  • Tool Calls: every tool it called, with the result
  • Risk State: exposure, executor count and drawdown at the time
The Runs tab of the Market Making Expert, showing a dry run's agent response
Refused actions are written to the session’s journal with the reason, so you can see why the risk gate blocked a call.

Notifications

Condor tells you about a loop when something needs your attention:
  • a tick error (once when it starts or changes, not on every tick)
  • a tick skipped because a risk limit was breached
  • a Dry Run or Run Once finishing or failing
  • a loop reaching its max_ticks limit
  • an emergency shutdown
These appear in the notification bell in the Workspace’s top bar, and in Telegram if you use it. Each one links to the agent’s page.

From Telegram

Telegram has no buttons for loops. Talk to the agent instead:
  1. Send /agent, tap Talk to and choose the agent.
  2. Ask in plain words. For example: “Is your loop running?”, “Pause your loop”, “Stop your loop” or “Shut your loop down and close the positions”.
The agent uses its control_agent tool, which can list, start, stop, pause, resume and shutdown loops. Starting a loop asks you to approve it first.
/stop only cancels the reply the agent is writing. End session in the /agent menu ends your chat with the agent. Neither one stops a running loop.

Next step

Routines

Extend agents with deterministic Python code
You’ve got an agent running. Tell us what’s working and what Condor should do next.

Help shape Condor

What’s working? What’s confusing? What do you wish Condor could do? Your answers go straight to the Hummingbot Foundation team and directly shape what we build next.