FicelleStrict-zero routing: why your agent should never spend behind your back
Most "free LLM" setups have a silent paid fallback. Strict-zero removes that risk: free pricing required, tools required, and no paid fallback, full stop. Here is what that means and why it matters.
The fastest way to get burned by an “AI agent” is a silent paid fallback. The free model rate-limits, the framework quietly retries against a paid one, and you find out at the end of the month. Strict-zero is the opposite default.
What strict-zero means
By default, Ficelle routes only to models that satisfy three rules:
- Free pricing required. A model enters the routing pool only if it is genuinely free at the tier you’ve configured.
- Tools required. For agent work, a model must support tool calls, or it isn’t a candidate.
- No paid fallback. When every free candidate is exhausted, Ficelle fails the request rather than silently spending money.
allow_paid_fallback is hardcoded false. If configuration claims otherwise,
Ficelle refuses to start. Paid fallback is not an opt-in switch.
The anti-false-free guard
“Free” is not always free. Some providers bill the same model IDs the moment your account carries a budget; others advertise a free tier that quietly meters paid credits. Ficelle runs an anti-false-free guard: if the runtime reports a model as billed, that model is quarantined out of the free pool. The catalog claim (“this is free”) is kept separate from verified behaviour (“this actually billed nothing”).
Why a hard failure is the right default
A failed request is loud, debuggable, and free. A silent paid call is invisible until the invoice. For agentic workloads that fan out thousands of calls, “invisible” is the dangerous one. Strict-zero makes the safe outcome the default and the spend an explicit decision.
What it is not
Strict-zero is not “unlimited free AI”. Free pools depend on provider catalogs and quotas, and they run out. That is exactly why Ficelle pairs strict-zero with safe fallback, cooldowns, and observability, so when free capacity is gone, you see it and decide, instead of paying by accident.
No middleman, either: your keys stay on your machine and traffic goes straight to the provider you route to. Ficelle is a local router, not a hosted proxy.