FicelleFree models break.
Your agents shouldn't.
When a free provider rate-limits or errors, Ficelle reroutes to the next working free model in milliseconds. Paid is blocked by design. It can't spend a cent, even if it wanted to. Your agents stay up, your card stays untouched.
Free is easy. Free that works isn't.
One model dies, the next takes over.
Rate-limited, erroring, or lying about what it supports? Ficelle drops it, cools it down, and reroutes. Your agent never feels it.
It stays free, guaranteed.
Ficelle only routes to models that are genuinely free and can run your tools. Nothing else qualifies.
It tests models so you don't have to.
Ficelle checks which models really handle tool calls, JSON, and vision, then ranks them by real speed and reliability, live.
It literally can't charge you.
Paid fallback is hardcoded off: flip it on and Ficelle refuses to start. And the open core is inspectable, so you don't have to take our word for it.
allow_paid_fallback = False # hardcoded, not a setting config_store.py:39You are paying frontier prices for busywork.
Most agent traffic is labelling, extraction and glue. A free model handles that as well as the expensive one. Send it through Ficelle and keep your API budget for the work that actually needs it.
Labelling and triage
Tagging tickets, detecting intent, sorting inbound, picking a branch in an agent graph.
ficelle/auto-fastFree tied the premium model on every task we tested.Structured extraction
Pulling fields out of invoices, emails, scraped pages, logs. Anything that ends as JSON.
ficelle/auto-jsonNear parity, and this is the workload agents run most.Reasoning and code
Multi-step logic, small refactors, test scaffolding, working through a plan.
ficelle/auto-reasoningThe one we expected free to fail. It held up.Polished writing
Launch copy, customer emails, release notes. Prose someone will judge you on.
keep your paid modelPaid genuinely writes better here. We are not going to pretend otherwise.Verdicts come from a 76-task benchmark against gpt-5. See the numbers below →
http://127.0.0.1:8646/v1Up and running in three steps.
Point your tools at one endpoint
Claude Code, Cursor, Continue, Open WebUI, your own scripts. Anything that speaks OpenAI works. Just swap the base URL.
Ask for the kind of model you need
Call ficelle/auto-tools, auto-json, or auto-fast. You say what the job needs; Ficelle picks the real model that delivers it.
Ficelle keeps it alive
It scores every working free model, routes to the best, and reroutes the instant one breaks. No config, no babysitting.
It speaks OpenAI. Your stack already works.
One base URL, and the key stays a placeholder: your real provider keys live in Ficelle, on your machine. Nothing else in your code changes.
from openai import OpenAI
client = OpenAI(
base_url="http://127.0.0.1:8646/v1",
api_key="ficelle-local",
)
client.chat.completions.create(
model="ficelle/auto-tools",
messages=[{"role": "user", "content": "..."}],
)curl http://127.0.0.1:8646/v1/chat/completions \
-H 'Content-Type: application/json' \
-d '{
"model": "ficelle/auto-json",
"messages": [{"role": "user", "content": "..."}]
}'export OPENAI_BASE_URL="http://127.0.0.1:8646/v1"
export OPENAI_API_KEY="ficelle-local"
# Every OpenAI-compatible tool on this machine now
# routes through Ficelle. No code change.- Claude Code
- Cursor
- Codex
- Continue
- Open WebUI
- LangChain
- CrewAI
- Hermes
- and anything else that speaks the OpenAI API
We measured it: 76 real agent tasks, free vs gpt-5.
Long-form content is the honest exception: paid still writes better. The whole 76-task benchmark cost $0.68 to run. Read the method and the raw numbers →
The router is free. Pro does the upkeep for you.
The core router and one-command install are free. Pro maintains the broader provider pool and adds fusion, prompt compression, one-command updates and support.
Open core (BSL 1.1). No account, no card.
InstallBilled monthly. Cancel anytime.89 € billed once a year.
Start 14-day trialOne payment. No renewals, ever.
Become a Founder14-day free trial, card required. Cancel anytime before it ends. The199 € Founder license is capped to the first 200. Routing stays local; Pro periodically reconnects to refresh its entitlement.
The things people ask before installing.
Is it really free?
The open core is free under BSL 1.1. It includes strict-zero routing, OpenRouter and Nous reference providers, and the one-command installer. Pro adds the maintained provider pool, model fusion, prompt compression and one-command updates. You never pay for models through Ficelle, on any plan.
What happens when a provider rate-limits?
Ficelle marks the model as cooling down, reroutes the request to the best working free model, and probes the limited one until it recovers. Your agent sees a normal response, not a 429.
Are free models good enough for real work?
For labelling, triage, structured extraction and most reasoning: yes. We benchmarked 76 real agent tasks against gpt-5 and free routing was quality-safe on classification (100%), JSON extraction (95%) and reasoning (94%). Polished long-form writing is the exception, where the paid model still wins. Route the first three through Ficelle and keep your budget for the writing.
Do my prompts or keys pass through your servers?
No. There is no middleman: the router runs on your machine, your API keys stay local, and prompts go straight to the provider you choose. The only network call Ficelle itself makes is the Pro license check, which carries license identity and nothing else.
Can it accidentally spend my money?
No. Paid fallback is hardcoded off. It is not a setting you can mis-toggle, and Ficelle refuses to start if the config claims otherwise. Any model that reports billing gets quarantined automatically. The core code is public to inspect, so you can verify all of it.
Does the Pro trial need a card?
Yes. The 14-day trial asks for a card but charges nothing until day 14, and you can cancel anytime before that. Annual (89 €/yr) is the best price; Lifetime (199 €) is a single payment, no renewals. Prices are in euros: Stripe charges the EUR amount and your bank converts it at your card's usual rate.
Which tools work with it?
Anything that speaks the OpenAI API: Claude Code, Cursor, Codex, Continue, Open WebUI, Hermes, LangChain, CrewAI, plain curl. Point the base URL at http://127.0.0.1:8646/v1 and keep your existing code.
Keep your agents running.
One command. No account, no card, and no way for it to charge you.
curl -fsSL https://raw.githubusercontent.com/TheBlueHouse75/ficelle-open-core/v0.1.6/scripts/bootstrap-ficelle.py | python3 -