Your agents run your Loops: a Loop is a definition, not a service
A Loop in Oddlynew is a definition: what the recurring work is for, what one cycle does, when it is due, and a ledger of every cycle that ran. Oddlynew keeps the definition and tells agents which Loops are due. It does not run them. Your agents do, the ones you already use, with the compute and the cadence you choose.
Why it works this way
You choose the agent. A frequent, simple check can run on a light, cheap agent; a weekly review on a stronger one. Different Loops can run on different agents.
Nothing spends on your behalf in the background. No hidden worker runs your Loops; every cycle is an agent you set up, using your plan or your keys.
Every cycle leaves a record. Each run is admitted in the Loop's ledger before it acts and closed with what it did, so the next run, and any person, sees what happened.
Two ways a Loop runs
While you work. A connected agent working in your organization checks which Loops are due (
loops.listDue). It runs a due Loop that takes no outside action and tells you in one line; a Loop that would send, publish, pay or change something outside waits for your yes.On a schedule. Any agent that can run on its own schedule can run a Loop for you: scheduled tasks in your chat client or coding agent, a scheduled job in your CI with an organization API key, or any always-on agent you use. Setting it up is one sentence to that agent.
Set one up
Copy the Loop's link from the web app (or ask your agent for it), then tell the agent that should run it:
Every weekday at 8:00, run the Loop at [link to the Loop] in our Oddlynew organization: read its definition, do one cycle, record it in its ledger, and tell me only what needs me.
Change the time and the last sentence to suit the Loop. To run it once now, leave out the schedule. To stop it, tell the agent, or pause the Loop in Oddlynew so every agent sees it is paused.
Boundaries
A Loop is not a permission. Running it never grants the agent more than it already has; outside actions still need you, or the approval rules of your organization's connections.
Keep one agent per Loop schedule. Two agents on the same schedule would both find it due; the ledger shows the second that a cycle is already running, but one owner is simpler.
A Loop with no runner does nothing. If nobody is running a Loop, it stays due, and the next agent that works in your organization will see it.
Related
How the goal, the work and the evidence of a Loop fit together: Goals Give Loops Direction. Choosing agents and cadence for upkeep: How to Start, section "You choose how gardening runs".