Triggered
A routine kicks off with no one waiting.
A scheduled task or event trigger starts the work. There's no chat window open and no spinner — nobody's staring at the first token.
How the bill works
Your bots don’t need to be fast.They need to be cheap and right.
A named teammate’s scheduled and background work isn’t like chat — nobody’s watching it happen. Flex Mode is how Chatticus is being built to notice that difference and spend it: work that isn’t due for hours or days can run through a slower, cheaper lane on the exact same model, instead of paying for interactive speed it doesn’t need.
“Do not replace the model with a cheaper brain. Give the same brain more time.”
From the design memo — see the full argument below
How it works
Queued with patience
If the deadline allows it, it waits in the cheap lane.
The same request goes out on OpenAI or Anthropic's Batch or Flex lane instead of the interactive one — slower to start, priced for patience.
Delivered on schedule
The result lands before the deadline.
Same model, same prompt, same tools — just paid for like scheduled work instead of a live conversation.
The two numbers
What this actually saves
OpenAI and Anthropic both discount their Batch and Flex lanes by about 50% — but that’s 50% off eligible token spend specifically, not 50% off the whole bot farm’s bill. Total savings depend on how much of your spend is inference in the first place, and how much of that inference can actually wait.
A worked example
60% inference × 70% eligible × 50% off = 21% total savings.
Still real money — just not the number you’d get by rounding “up to 50% off” and applying it to the whole bill.
How the lane gets chosen
Scheduled work, priced for patience
Work moves through queues
A named teammate's background work runs asynchronously — turns move through queues and journals with server-sent progress, not a live socket waiting on a reply.
Costed against real numbers
The cost math and provider facts — OpenAI and Anthropic's Batch pricing, rate limits, and completion windows, checked against their current docs and worked through with real numbers.
The scheduler picks the lane
The scheduler decides, per task, whether there's enough slack before the deadline to route it through Batch or Flex instead of the interactive API.