Budgets and a kill switch for your AI agents.
Give each person, agent, workflow and key its own budget for AI. Every call is checked before it runs: a warning near the limit, a stop at it. Then see what each call cost, and how to do the same work for less.
One Programmable account across the suite.
Early access: you can set budgets and see your spend today. Budgets that stop calls are being switched on across the platform.
- Checked before every call
- Warns at 80%, stops at the cap
- Fails closed if a check can't run
- Your provider keys stay yours
Most AI bills arrive as one number, a month late.
Spend splits the bill by who and what made each call, and stops it before it grows. These are the bills it's for.
A loop that won't end
An agent retries the same expensive call all night. It stops at its daily budget, not at your card's limit.
One busy afternoon
Each person has their own budget, so one person's big day can't use up the team's month.
A key that runs away
A key you gave to an app or a customer carries its own dollar limit, and stops there.
A big model on a small job
Spend finds the calls a smaller model does just as well, and shows what switching would save.
A budget for
everything that spends.
Set a daily or monthly amount for a person, an agent, a workflow, a workspace or a key. Each one shows where it stands today, and warns you at the level you choose.
- Per person
- Per agent
- Per workflow
- Per key
- MayaPerson$120 a month$55.20 · 46%
- Research agentAgent$20 a day$12.40 · 62%
- Weekly reportWorkflow$40 a month$33.60 · 84%
- Acme app keyKey$250 a month$45.00 · 18%
Over budget means stopped.
Not billed.
Every call is checked before it runs. One that would go over a budget never runs, and your agent is told why. Pause all spending in one click when something looks wrong.
- Checked before it runs
- Fails closed
- Pause everything
Every reason a call can be stopped is on Why a call was stopped, and whether checks are working is on Status.
- Summarize a report · Research agent$0.42
- Draft a reply · Maya$0.18
- Compare two contracts · Research agent$1.62
Summarize a report · Research agent. It would have gone past today's $20.00, so it never ran, and the agent was told why.
Why a call was stoppedSame work.
A smaller bill.
Spend looks at your own calls for jobs a smaller model does just as well, and shows what it checked and what you'd save. Switch one on and it becomes a rule; until then nothing changes.
- From your own calls
- Shows its working
- Your call
- Calls checked
- 212
- Same result
- 209
- Cost
- −80%
Three steps, then every call is checked.
No new account and no new keys to hand over.
- Step 1
Connect your agents
Use your programmable.run keys as they are, or point an app at the Spend endpoint, SDK or proxy. Your provider keys stay where they are.
- Step 2
Set budgets
A daily or monthly amount for a person, agent, workflow, workspace or key, and the level where you want a warning.
- Step 3
Every call is checked first
Allowed, warned or stopped, before the call runs. Then you see what it cost and what it got done.
Works with the agents you already have
Change one line in an app that uses the OpenAI or Anthropic SDK, and its calls go through your budgets on your own provider key. Or wrap calls with the SDK, or run the proxy next to your app. The app shows which of these are switched on for your account. Read the docs
client = OpenAI( base_url="https://llm.programmable.run/v1", api_key=SPEND_KEY, # a key with its own budget )
Built to stop, not just to report.
Dashboards tell you after the money is gone. Spend checks first, and every rule leans toward stopping.
A stop is a stop
When a call would go over budget, it doesn't run. If the check itself can't be made, the call doesn't run either.
Warnings before the end
A warning when a budget reaches the level you choose, 80% unless you change it, before anything is stopped.
Your keys stay yours
Provider keys stay in the Programmable vault. Spend never stores them and never shows them to anyone.
Every call accounted for
Cost and outcome for each call, and who it was for. Spend keeps the numbers, not your prompts.
Cheaper routes, your call
Suggestions come from your own calls. Nothing changes until you switch one on, and you can switch it off.
Raising a limit asks first
Raising or removing a budget asks you to confirm, so a limit is never loosened by accident.
Good questions, straight answers.
What happens when a budget is reached?
Budgets that stop calls are being switched on across the platform. Once they are, the call is stopped before it runs, so nothing is spent. Your agent gets a clear refusal with the reason, and you see the stopped call in your log. Every reason is explained on Why a call was stopped.
What if Spend can't check a call?
Once budgets stop calls, the call doesn't run. Checks fail closed: a call is never let through unchecked. You can see whether checks are working on the status page.
What does it cost?
Spend is included with programmable.run, on the same Programmable account. The model calls themselves are paid as they are now: by your provider on your own key, or with credits on the platform's key. Each call says which.
Do you see my prompts or my provider keys?
No. Provider keys stay in the Programmable vault, and Spend never stores or shows them. For each call Spend keeps the numbers: the model, the cost, the outcome and who it was for.
Which agents does it work with?
Calls made with programmable.run keys, and apps that use the OpenAI or Anthropic SDK, through the Spend endpoint, SDK or proxy. Parts switch on as they go live, and the app says plainly which ones work for you. The docs show how to connect each, and what's new lists what just arrived.
Will a cheaper route change my agents without asking?
Never. Suggestions come from your own calls and show what they're based on. A suggestion becomes a rule only when you switch it on, and you can switch it off again.
Set a budget before the next bill.
Sign in with your Programmable account. Spend is included with programmable.run.
Docs · Status · What's new