Spend caps and alerts
Set a monthly AI budget. Get alerts at 50, 75, and 100%. Auto-pause at 100% so you don't get surprised.
Spend caps and alerts
A monthly cap on AI spend, with alerts on the way up and an auto-pause at 100%. This is your seatbelt against the rare misbehaving channel that floods you with bot traffic, or a launch day that catches you off guard.
What you set
Three things on AI → Spend:
- Monthly cap. A dollar amount.
- Alert thresholds. 50%, 75%, and 100% (defaults). All three notify Owners and Admins.
- Auto-pause at 100%. Default on. AI replies stop until next billing cycle or you raise the cap.
How spend is counted
Every AI reply has a cost in AI receipts. The spend cap is the running sum of those costs in the current calendar month, in your billing currency.
It includes:
- Drafting tokens (input + output).
- Auto-labeling tokens.
- Embedding generation for new KB articles and re-scans.
It does not include:
- Failed requests where the provider returned an error before billing.
- Anything inside the Playground (Playground spend is tracked but not capped).
What happens at each threshold
- 50%. Email to Owners and Admins. "You are halfway through your AI budget."
- 75%. Same email. "You are at 75%."
- 100%. Auto-pause kicks in. AI replies stop. Inbox keeps working. Owners and Admins are emailed.
You can disable any threshold or the auto-pause itself.
What pauses, what doesn't
When auto-pause fires:
- AI drafting: paused.
- AI auto-send: paused.
- AI auto-labeling: paused.
- KB embedding generation: paused.
What keeps working:
- The inbox.
- Routing and SLA rules (without AI labels).
- All human reply flows.
- Receipts (so you can audit what got spent).
Raising the cap
Owners and Admins can raise the cap in real time. The auto-pause clears the moment the new cap is above the current month's spend.
If you raise it during a 100% pause, AI replies resume on the next message. In-flight conversations are not retroactively drafted.
Cap math you should know
Per-resolution cost typically lands between $0.005 and $0.01 on Haiku 4.5 or GPT-4o-mini, and proportionally more on the flagships. See Choosing a model. For a workspace doing 10,000 AI resolutions a month on Sonnet 4.6, expect roughly $100 to $250 in AI spend.
A reasonable first cap:
- Estimate volume per month from your inbox stats.
- Multiply by $0.015 (a generous per-resolution number).
- Double it.
That gives you a cap you are unlikely to hit through normal use, but you will hit if a chatbot or scraper starts hammering your support address.
Watching spend over time
The Spend page shows daily spend and a trend chart. Patterns to look for:
- A new channel adding more cost than expected.
- A specific model running hot.
- A spike correlating with a specific topic (often a misconfigured automated system spamming your inbox).
Cap vs provider limits
The Ochre cap is independent of your Anthropic or OpenAI account limits. You can have:
- A $5,000 limit at your provider, $1,000 at Ochre.
- A $100 limit at your provider, $1,000 at Ochre.
The lower of the two is what actually applies. Many teams set the Ochre cap as the soft limit and the provider limit as the hard limit, with a buffer between them.
Cap and BYOK
Spend caps work the same whether you use one BYOK key or two with fallback. Cap is in dollars, regardless of which provider produced the reply.
Notifications
Cap alerts go to Owners and Admins by email. You can also pipe them to a Slack channel via the Slack integration.
Recommended setup
- Monthly cap: 2x expected usage.
- Alerts at 50, 75, 100%: all on.
- Auto-pause at 100%: on.
- Slack alert channel set if you have one.
- Re-evaluate after the first month.
Related
Was this article helpful?