Daily AI safety limit

The daily AI safety limit is a safeguard for your whole organization. It caps how much AI your organization uses in one day, so a runaway agent, or a loop that keeps asking AI for more, can’t cause an unexpected spike. When today’s AI use reaches the limit, AI pauses for everyone until the day resets at 00:00 UTC, or sooner if an admin raises the limit.

This is a safeguard, not a billing problem. When it kicks in, nothing is wrong with your account or billing. It is the limit doing its job.

How it works

One day at a time. Each day starts at zero at 00:00 UTC. Ronja always shows that moment in your own time: east of UTC, as in Stockholm, it falls in the small hours, so for most of your day it reads “tomorrow”, for example “AI resumes by tomorrow at 02:00.”; west of UTC, as in New York, it falls the same evening. Most examples below show Stockholm between midnight and the reset, when it reads “by 02:00”.
All AI counts, for everyone. Chats with Ronja, Slack and Teams conversations, Saved Agents, automations’ Agents, and apps and workflows that ask AI something all count, admins included.
Two limits, the lower one applies. Ronja’s limit for your organization is set by Ronja. Any Admin can also set your organization’s daily AI safety limit, at or below Ronja’s, to pause sooner. Whichever is lower is the one in effect.
A heads-up first. At 80%, your admins are told and everyone sees a note above the chat box. At 100%, AI pauses.

What you see when AI is paused

A grey banner above the chat box explains it:

AI is paused for now. Your organization’s daily AI safety limit kicked in. It protects you from an unexpected spike in AI use. Nothing is wrong with your account or billing. AI resumes by 02:00.

The time is in your own time zone, and it says “by” because an admin can lift the pause sooner. Before that, from 80%, the banner reads, for example, “Your organization has used 85% of today’s AI safety limit. If it’s reached, AI pauses until tomorrow at 02:00.”, as it reads in Stockholm during the day. Members only ever see this percentage, never the limit itself.

If Ronja briefly can’t check today’s use, the banner may not show even though AI is paused. A message you send then can still come back with the usual “Not sent — AI is paused.” line.

While AI is paused:

A message you send stays in the chat with a grey line under it: “Not sent — AI is paused. It resumes by 02:00.” Click Retry once AI resumes, or Discard to drop it. Nothing you typed is lost.
A queued prompt reads Paused and stays queued. Run it once AI resumes.
A reply already under way stops before Ronja’s next step, and she says that AI is paused.
Approve on a step, Skip for now on a form and submitting a form are refused, and the step or form keeps waiting for you. Reject still works.
In Slack or Microsoft Teams, Ronja answers each message with the same explanation instead of a reply.
An answer to a question an Agent or automation asked you isn’t kept. Answered in a chat, it is refused and the question keeps waiting. Answered by email, Slack or Teams, Ronja writes back “Thanks, but I can’t carry on right now.” with the same explanation. Reply again once AI resumes. If the time to answer has run out by then, the run carries on as it does when nobody answers.

What keeps running and what pauses

AI pauses for all your organization’s work: chats in every mode, Slack and Teams conversations, Saved Agents however they are started (from a chat, an automation, a workflow, an app or the API), and any app or workflow step that asks AI something. Only Ronja’s own background work (naming chats, summaries, scanning tables for personal data, indexing documents) keeps running, and it still counts.

Everything else keeps running. Workflows that don’t use AI run as usual, including the ones automations start, and so do table builds, connection syncs and queries.

A run that is waiting on a person, an approval or a set time keeps waiting, and carries on once AI resumes.

Automations during a pause

An automation whose action is a Saved Agent doesn’t run while AI is paused. Each fire is recorded in its run history as a skipped run labelled Paused — daily AI safety limit, in neutral grey. It isn’t a failure: it doesn’t count towards a run of failures and sends no failure alert. A run the pause stopped part-way through is recorded the same way.

An automation whose action is a workflow still starts it, and a workflow that doesn’t use AI runs as usual. When one of its AI steps is refused because AI is paused and the workflow fails on it, the automation’s run is recorded the same way: Paused — daily AI safety limit, its summary the same explanation, with no failure alert, and it doesn’t count towards a run of failures. The workflow run it links to reads Failed, with the same explanation as its error, and its log ends with the script’s own error line. A workflow written to catch the refusal and carry on completes as usual when it finishes; if it fails later in the same run, that run is recorded as paused too.

A fire during a pause is not run again later, whatever the trigger. A scheduled automation runs at its next scheduled time. An email, webhook, event, table change or mailbox message that arrived during the pause is not replayed, and what it carried is not kept. Run now starts the automation again without it once AI resumes, except for a workflow that takes its inputs from the email or event: with nothing to fill them, that run fails. For an email or webhook that mattered, have it sent again once AI resumes, or ask Ronja to handle it.

Workflows started another way

A workflow you ran by hand, or that an app or Ronja started, also fails when one of its AI steps is refused during a pause, with the same explanation as its error. It sends the usual Workflow run failed notification, and its email at most once a day for the same workflow (see When a workflow run fails). Run it again once AI resumes.

Over the API

A request that would use AI while AI is paused, such as starting a Saved Agent run or a workflow step asking AI something, answers 400 with the code daily_spend_limit_reached, the same explanation, resetsAt (when the day resets, in UTC) and owner: ceiling when your organization’s own limit is in effect, quota when Ronja’s is. A Saved Agent run already going ends failed with the failure class daily_spend_limit. A workflow run that fails on the pause ends error, with the explanation as its error. An automation’s paused run, whether it ran a Saved Agent or a workflow, is skipped with failureClass daily_spend_limit. Don’t retry before resetsAt unless an admin has raised the limit. See the API reference.

For admins

Every Admin sees today’s use and both limits, with or without billing access.

See today’s use

On the Control Center’s People tab, the Today line says how much of the limit the organization has used, for example “62% of the daily AI safety limit · resets at 02:00”. While AI is paused, a grey strip at the top of the tab says “AI is paused for everyone until 02:00 — daily AI safety limit.” and who can change the limit, members’ status reads Paused for today, and the Paused for today filter lists them. A member with no role shows No role instead, and everyone shows Credits ran out if the organization has used all its credits. Each person’s panel has a Daily AI safety limit section too.

Set your organization’s limit

Open the Control Center (admin sidebar → Billing & Control → Control Center) and click Rules.
Under Daily AI safety limit, click Set a limit, or Edit if your organization already has one.
Enter Credits per day, counted as on your bill. It must be at or below Ronja’s limit, which the dialog shows.
Click Save limit.

To go back to Ronja’s limit alone, click Remove, then Remove limit. A change takes effect within a minute, so raising the limit during a pause lifts it without waiting for 00:00 UTC. Setting the limit is browser-only: an API token, the CLI and automations can’t change it.

The section shows three rows: Your organization’s daily AI safety limit (or Not set), Ronja’s limit for your organization, and Today, for example “62% used · resets at 02:00”. In effect marks the limit that applies. To have Ronja’s limit raised, contact Ronja support: your own limit can’t go above it.

The daily limit counts credits as they appear on your bill, so its numbers aren’t directly comparable with a weekly AI budget’s. Use can end slightly above 100%: Ronja checks the limit before every step, from a reading at most about half a minute old, so steps already under way finish, and her own background work keeps running and still counts (see What keeps running and what pauses). If Ronja briefly can’t check today’s use, she keeps the day’s last reading for longer: a pause stays in place, but use can run further past 100%.

If the line Pausing is switched off by Ronja right now appears, Ronja has turned pausing off for the moment: today’s use still shows, but reaching the limit pauses nobody and sends no notice.

Get told before and when it happens

At 80% or more of today’s limit, and again when it is reached, your admins get an email, a notification in Ronja and, if your organization routes it there, a Slack message. Each says how much is used, never an amount, and who can change the limit. You hear about each at most once a day, and again if the limit changes. These notices are on by default; to send them somewhere else or turn them off, use the Daily AI safety limit row in the AI safety section of Notification routing (admin sidebar → Governance).

When other limits apply too

If your organization has used all its credits, or its monthly budget, that message comes first. Otherwise the daily AI safety limit comes before a person’s own weekly AI budget or billing-period allocation. Unlike those, it pauses all AI, automations’ Agents included. See What each limit does.