FAQ

Does this replace my gateway (LiteLLM, etc.)?

No — it complements it. A gateway caps a key or org, enforced per request (a 429 when the key is over). Breakerbox caps one run and trips between steps. Because a key limit is set high enough not to block real work, a single runaway run overspends underneath it — the limit only fires once the damage is account-wide. Run both.

Do you see my prompts or my provider API keys?

No. Breakerbox never stores, receives, or transmits your LLM/provider keys. Bring-your-own-key features (like AI code suggestions in the builder) call the provider directly from your browser. See the security model.

Does it run my graph on your servers?

Never. The visual builder is codegen-only — it generates plain Python you download and run yourself. There's no server that executes your flow, so there's nothing to hack into.

What frameworks does it support?

Today guard() wraps compiled LangGraph apps. Adapters for other agent frameworks are on the roadmap.

How is cost metered — is it accurate?

Tokens are self-metered with tiktoken, then reconciled against the provider's reported usage, so the receipt reflects what actually happened. Prices come from a built-in table; unknown_model controls what happens for models it doesn't recognize (see the reference).

What happens to side effects when it trips?

It trips at hop boundaries, never mid-call, so no request is interrupted. On a kill, the receipt lists any side-effecting tools that fired before the stop.

Can I resume a paused run?

Yes. on_trip="pause" checkpoints the run so you can resume it with more budget (see topup_policy in the reference).

Is the cloud dashboard required?

No. Everything works with zero cloud — a terminal summary and a single-file HTML receipt on your machine. The dashboard is optional, for a shared live view of runs across a team.

Is it really open source?

Yes — MIT, and the library stays MIT. See LICENSE.