Questions and answers

Short answers, each with a link to the page that has the detail.

What is a spend limit for an agent?

A budget that an agent, a run or a task can spend from and cannot go past. In buckette it is a bucket: one or more balances, in tokens, dollars, calls or any unit you name. Your code asks buckette before it spends, and the spend is taken from the bucket only if it fits. A bucket can be split into child buckets, one per agent or per run, each with a key that cannot touch anything outside it. Getting started shows the first steps.

How is this different from my model provider’s own spending cap?

A provider’s cap covers what you spend with that provider, cut the way the provider chooses. buckette’s limits are cut the way you choose: per agent, per run, per customer or per workflow, across several providers, in whatever unit you count. They refill on a schedule, warn you as a balance crosses the levels you set, and leftovers can return to the budget they came from. buckette meters what you report to it: it cannot see a provider’s own counter and does not change what a provider charges. buckette for token-maxxers works through one case.

What happens when a budget runs out?

The spend is refused and nothing is taken, so your code can stop, or wait until the next refill (buckette can estimate when that will be). You can also ask for the request to be queued and served when room comes back. Refused spends are recorded, so you can see the demand a budget did not cover, and webhooks warn you before a balance reaches zero. buckette for workflows has more.

How do I stop an agent that is spending too fast?

Freeze its bucket or revoke its key, from the Console or the API. A freeze refuses every new spend from the bucket and from the buckets under it, until you unfreeze them. A revoked key stops working for whoever holds it. Through the API you can also reclaim a child bucket when its work ends: what it did not use returns to its parent, and the keys made for it are turned off.

Does buckette see my prompts or what my agents do?

No. buckette receives the amounts you charge against a bucket and the names and metadata you choose to give buckets. The integration libraries for agent frameworks send token and tool-call counts, never the text. Your data lists what is stored and what is not.

Where is my data hosted?

On Google Cloud, in the europe-west1 region (Belgium, in the European Union), with backups in the same region. You can export your buckets and their history, and delete a tenant or your account, at any time. Your data has the detail, and the terms of service are the binding text.

Does my code have to change?

Your code, or your agent’s, makes an HTTP call to buckette before it spends and, for a hold, another when the work ends. Any language that can make an HTTP request can do it. A coding agent can also set you up on its own from one sentence: Read https://console-api.buckette.co/docs/llm.md and set me up on buckette. Getting started describes both routes.

Can I run buckette on my own machines?

Yes, on request. The buckette API and a Console for it can run on your own machines as container images, with your own PostgreSQL database. There is no public price for this; write to hello@buckette.co.

What does it cost?

An account may hold one free tenant. Beyond it, a tenant uses a slot on a plan, at a monthly price per slot, or is switched to the Elastic plan, billed each month for the buckets it holds. The plans and their prices are listed in the Console when you buy; this site does not quote them yet. Pricing explains how the plans work, and the payment terms say how you are billed.