AI credits and limits
How Halo AI's use is counted, what each plan includes, the three ways to pay, and every cap that can stop an answer.
AI credits measure what Halo AI costs. Every call counts by what it costs on the model that ran it, so a quick description uses a few credits and a full report uses thousands.
What a credit is
One AI credit is a hundredth of a US cent of Claude's list price. Each call is rounded up to whole credits. Faster models use fewer credits for the same work, because they cost less.
| What | Model | About |
|---|---|---|
| A name or a field description | Haiku 5.5 | 1–20 credits |
| A chat answer with two or three queries | Sonnet 5.5 | 200–1,500 credits |
| A full report plan | Sonnet 5.5 | 2,000–8,000 credits |
| The same on Deep | Opus 5.5 | About twice Sonnet |
What each plan includes
Halo AI comes with every plan, with a monthly amount of credits. Halo can set a different amount for your organization, and Settings, Usage shows yours, as AI credits per month.
| Plan | Credits a month | Value at list price |
|---|---|---|
| Standard | 200,000 | $20 |
| Professional | 1,000,000 | $100 |
| Enterprise | 5,000,000 | $500 |
Months count in your organization's time zone. Trials get their plan's credits.
Three ways to pay
Halo's team sets which one your organization uses. Usage and billing in Settings, Halo AI shows which. Ask your account manager to change it.
| What happens | Who pays Anthropic | |
|---|---|---|
| Included with your plan | Your plan's credits. When they're used up, Halo AI stops until the month turns or Halo adds more | Halo |
| Included, then pay as you go | Your plan's credits first, then pay-as-you-go credits up to a monthly cap you've agreed with Halo. They're billed on the month's invoice | Halo, billed on to you |
| Your own Anthropic key | Claude runs on your organization's Anthropic API key. Halo's credits aren't used, and Anthropic bills you directly | You |
Pay as you go
Pay-as-you-go credits cost Claude's list price plus Halo's markup, which is 25% unless Halo has agreed another rate with you. So 1,000 pay-as-you-go credits cost 12.5 cents against 10 at list price, rounded up to the cent on the month. Usage and billing shows your rate and your cap. Past months list their credits and amount.
Your own Anthropic key
Any plan can use its own key, with no platform fee. Once Halo has set your organization up to pay this way, Your Anthropic key appears under Usage and billing, for people who can manage organization settings.
Make a key in the Claude Console, for a workspace with access to Claude's models.
Paste it under API key and press Save key.
Halo checks the key works before it stores it, and only its last characters are ever shown. Until a key is there, Halo AI is off for your organization. Replace swaps it. Remove key stops Halo AI for everyone until you add another. Conversations are kept either way.
The usage tables still show what your calls cost at list price, but nothing's charged. Halo's daily cap per member doesn't apply on your own key unless Halo sets one.
See what's used
Under Usage and billing in Settings, Halo AI:
Plan credits this month and Pay-as-you-go credits this month, as meters.
This month by feature: each feature and model, with calls and credits.
Past months: credits, pay as you go and amount for months that have closed.
This tab needs the See usage and limits permission. Anyone using Halo AI sees a smaller version in the Halo AI panel's footer: the limit they're closest to, such as "12% of credits".
Caps for members
Under Members in Settings, Halo AI:
Credits each member may use a month sets a cap for everyone. 0 means no cap.
Use and caps lists who used Halo AI this month. Open someone to give them Their own monthly cap, or to Pause Halo AI for them.
A pause stops them in this organization until you resume it. The reason you give isn't shown to them.
Every limit
Each check runs before a call, and again between an answer's steps.
| Limit | Default | Who sets it |
|---|---|---|
| Your organization's plan credits a month | By plan | Halo |
| Your pay-as-you-go cap a month | Agreed with Halo | Halo |
| Credits a member may use a month | No cap | Your Admins, with a per-member override |
| Credits a member may use a day | 50,000 | Halo |
| Credits one answer may use, across its steps | 20,000 | Halo |
| Answers being written at once | 2 per member, 10 per organization | Fixed |
| Helper calls at once: names, descriptions, formulas, SQL | 3 per member | Fixed |
| Steps in one answer | 12. A report from Generate with Halo or Rebuild with Halo can take 40 | Fixed |
| Text attached to one message | 120,000 characters | Fixed |
| Questions in one conversation | 60, then you ask in a new one | Fixed |
| Requests a minute | 40 per member, 400 per organization, and 30 new conversations per member | Fixed |
What happens at a limit
Near and at your plan's credits. Owners and Admins get a notification at 80% and at 100%. See Usage and plan limits.
At a limit. New questions stop. The panel says "Out of AI credits for now", with a way to Halo AI settings for people who can change them. Other features of Halo keep working.
During an answer. A call already running finishes and is counted. The counter stops at the cap, so the next call is refused.
A member at a cap. Each cap reached is flagged under To review, once a day if it keeps happening.
Stopping early isn't free
A stopped answer reports almost no use, since usage arrives at the end of a response. So Halo counts the larger of what was written and what the time spent writing allows. That covers Halo AI's thinking, which you don't see.
Switching features off
To spend less, an Admin can switch off any Halo AI feature in Settings → Halo AI, under Features, and see what each used this month. A feature that's off is off for everyone in the organization, and its buttons go away.
Most features only use credits when someone asks. Halo CMS's Generate websites writes a whole website, so each one uses a lot, and Halo CRM's Read with Halo uses a few for each read. Four features don't wait to be asked:
Catalog descriptions runs by itself after each discovery, and uses credits for each table or column with no description, up to 100 tables a run. Switch it off if you'd rather write your own.
Insights and written summaries runs by itself when a measure someone chose to watch changes, and uses credits for each change it explains, up to 10 a check. It also writes the summaries scheduled emails ask for, using credits for each send. Checking for changes uses none, so with this off they're still flagged and emails still send, just without those words.
Alert explanations runs by itself each time a data alert fires, and uses credits for each one it writes. Alerts still fire and send with it off, without those words.
Names titles each conversation after its first answer, at a very small cost. Off, a conversation takes the first words of its question.
Keep costs down
Leave Allow Deep mode off unless people need it. It's off for new organizations.
Leave the default mode on Auto, which uses Haiku wherever that's enough.
Set Credits each member may use a month if a few people use most of your credits.
Add your terms and definitions to What Halo AI should know, so Halo AI spends fewer steps working them out.
Members' own AI tools don't use your credits, because they run on their own AI provider. See The Halo MCP server.