Cost control
Know what every AI job costs.
Give every team an allowance that stops spend before it happens. See what each finished job cost. Get one statement a month that your CFO will actually read.
Every team gets an allowance
Hard caps in Australian dollars, enforced before a request runs.
Every dollar gets a receipt
One header turns token spend into cost per finished job.
One statement a month
What you spent, what it bought, and what it would cost by hand.
1 · Allowances
A cap that can’t be crossed.
Split your account into workspaces: a department, a project, or a client. Give each one a monthly allowance in Australian dollars.
Most AI services tell you what you spent after you spent it. A runaway script or a leaked key can burn a month’s budget overnight.
Ours works the other way. Every request reserves its worst-case cost before it runs. If it won’t fit, it is refused. Requests sent at the same moment can’t slip past. Try it.
Engineering workspace · sample data
A test script goes wrong and sends 80 requests in a row.
0
requests served
0
refused at the cap
A$0.00
over the allowance
Each request reserves its worst-case cost before it runs. When the next one won’t fit, it is refused. The allowance cannot be crossed.
Wattle Health Group · Workspaces · October 2026
ExampleCustomer service
At the cap: cheaper model, then stop
A$396.54 / A$480 · 83%
Finance
At the cap: stop
A$38.58 / A$600 · 6%
Clinical governance
At the cap: stop
A$12.43 / A$400 · 3%
Engineering
At the cap: cheaper model, then stop
A$1,880.50 / A$2,000 · 94%
Alerts at 50%, 80% and 100% of every allowance, by email, Teams or Slack.
2 · Receipts
Cost per job, not per token.
Add one header to name the workflow and the job. That’s the whole integration.
The console then shows what each finished job cost. Not tokens per day. Invoice checks at less than a cent each. A code review at sixty cents.
Before a workflow goes live, declare how long the job takes by hand. Every statement then measures against what you declared, not a number worked out afterwards.
from openai import OpenAI
client = OpenAI(base_url="https://api.continuumlabs.ai/v1", api_key="cl-...")
reply = client.chat.completions.create(
model="glm-5.2",
messages=[{"role": "user", "content": invoice_text}],
extra_headers={
"X-Continuum-Workflow": "invoice-check",
"X-Continuum-Outcome": "INV-2291",
},
)Cost per job · October 2026
Example| Workflow | Jobs | Spend | Each |
|---|---|---|---|
| Code reviewEngineering | 1,420 | A$852.00 | A$0.60 |
| Test evaluationsEngineering | 46,000 | A$690.00 | A$0.015 |
| Support ticket repliesCustomer service | 9,560 | A$286.80 | A$0.030 |
| Patient letter rewritesCustomer service | 1,840 | A$47.84 | A$0.026 |
| Invoice checksFinance | 3,214 | A$12.86 | A$0.004 |
| Incident triageClinical governance | 212 | A$9.33 | A$0.044 |
| Supplier cost reviewsFinance | 18 | A$4.32 | A$0.24 |
| Not labelled | A$424.90 |
3 · The statement
One page, once a month.
On the first of every month, each workspace owner gets one page. What was spent. What it bought. What’s coming. What looked odd.
And the number the board wants: what the same work would have cost by hand, against the time you declared for each job.
It prints to PDF. It downloads in FOCUS, the FinOps Foundation’s standard format, so finance can load it into the tools they already use.
AI statement · October 2026
Wattle Health Group
- Spent
- A$2,328
- 67% of allowances
- Jobs finished
- 62,264
- 82% of spend labelled
- Same work by hand
- A$91,602
- at A$45 an hour, declared
- November forecast
- A$2,607
- on current trend
| Workflow | Cost | By hand |
|---|---|---|
| Code review | A$852 | A$15,975 |
| Test evaluations | A$690 | — |
| Support ticket replies | A$287 | A$43,020 |
| Patient letter rewrites | A$48 | A$16,560 |
| Invoice checks | A$13 | A$9,642 |
| Incident triage | A$9 | A$3,975 |
| Supplier cost reviews | A$4 | A$2,430 |
Worth a look
- The Engineering key "ci-eval" tripled its daily spend on 14 October. Someone added a new test suite.
- Customer service passed 80% of its allowance on 24 October, six days earlier than in September.
By-hand figures use the minutes per job declared by each workflow owner before it went live. Test evaluations have no by-hand equivalent and are excluded.
For MSPs
Resell it as yours.
One partner account. One workspace per client. One wholesale invoice from us.
You set each client’s price. They see your prices, your logo and your statement. Never ours.
A new client takes five minutes: create a workspace, set an allowance, hand over a key. Or let your own systems do it through the management API.
Southern Cross IT · Clients · October 2026
Example| Client | Wholesale | Your price | Your margin |
|---|---|---|---|
| Wattle Health Group | A$2,328.05 | A$3,026.47 +30% | A$698.42 |
| Harbour & Co Accounting | A$412.60 | A$577.64 +40% | A$165.04 |
| Saltbush Design Studio | A$96.20 | A$134.68 +40% | A$38.48 |
| Banksia Legal | A$1,184.00 | A$1,598.40 +35% | A$414.40 |
| This month | A$4,020.85 | A$5,337.18 | A$1,316.34 |
How it works
Checked before it runs. Charged when it’s done.
- 01
Request arrives
With a key, and optionally a workflow and job.
- 02
Worst case reserved
In one database step, against the key, the workspace and the account. If any is short, it’s refused.
- 03
Model runs
In Australia. Prompts and answers are never stored.
- 04
Real cost settled
The unused reservation goes back. The job’s cost is recorded against its workflow.
The reservation step is already how every Continuum request is billed. Allowances extend it from the account to every workspace and key.
Start in five minutes.
- 1. In the console, open Cost control and add a workspace with an allowance.
- 2. Move your keys into it. Give any key its own limit if you like.
- 3. Add the
X-Continuum-Workflowheader to your requests, and declare what each job is worth. - 4. Your first statement arrives with the month. Watch it build in the console as you go.
Allowances come with every Continuum account. There is nothing extra to pay. The docs cover the details and the management API.
MSP or reseller?
Get a partner account: one workspace per client, your own prices, one wholesale invoice.