
What is Ficelle?
Ficelle is an OpenAI-compatible router that routes free LLMs for reliable inference. It solves the problem of free LLMs being unreliable and prone to rate-limiting or failures. Ficelle provides a single endpoint for 13 free LLM providers, with automatic failover in milliseconds.
- Website
- ficelle.ai
- Category
- AI & automation
- Pricing
- Free plan with paid upgrades
- Platforms
- web
- Compared with
- OpenAI, Anthropic
About Ficelle
Ficelle works by measuring, ranking, and rerouting free models to ensure that the user's agent keeps getting answers. It checks whether each model can handle tool calls, JSON, and vision, and ranks them by speed and reliability. Ficelle also has a feature called paid fallback, which is hardcoded off, so routed requests cannot incur paid-model charges.
Ficelle is suitable for developers and users who want reliable free LLM inference. It is offered in several plans, including a free plan with no account or card required, and Pro plans with additional features such as maintained provider pool, model fusion, and support. The Pro plans are priced at 12 EUR/month or 109 EUR/year, with a 14-day free trial available.
- The problem
- Free LLMs are unreliable due to rate limits and failures, causing applications to break.
- How it solves it
- Ficelle provides a single endpoint for multiple free LLM providers, automatically routing requests to the best available model.
- What makes it different
- Ficelle's open core is free under BSL 1.1, and its paid fallback is hardcoded off, ensuring that users are never charged for models.
Key features
- Automatic failover in milliseconds
- Paid fallback blocked in code
- One endpoint for 13 providers
- Free core with no account or card required
- Local routing with no middleman
- Model testing and ranking
Product details
Optional maker-provided information published with the launch. Confirm current details on the official website.
- Key features
- Automatic failover in milliseconds
- Paid fallback blocked in code
- One endpoint for 13 providers
- Free core with no account or card required
- Local routing with no middleman
- Model testing and ranking
- Pricing model
- Freemium
- Platforms
- Web
- Tech stack
- Not provided
How Ficelle works
- Point tools at one endpoint
- Ask for the kind of model needed
- Ficelle handles failover
Use cases
Labelling
Ficelle can be used for labelling tasks, such as tagging tickets or detecting intent.
Structured Extraction
Ficelle can be used for structured extraction tasks, such as pulling fields out of invoices or emails.
Reasoning and Code
Ficelle can be used for reasoning and code tasks, such as multi-step logic or small refactors.
Who Ficelle is for
- Developers
- Users who want reliable free LLM inference
- Those who need to route free LLMs for their agents
Questions about Ficelle
Is Ficelle really free?
Yes, the open core is free under BSL 1.1, with no account or card required.
What happens when a provider rate-limits?
Ficelle marks the model as cooling down, tries the best working free alternative, and probes the limited one until it recovers.
Are free models good enough for real work?
Yes, for labelling, triage, structured extraction, and most reasoning tasks, free models are quality-safe.
Do my prompts or keys pass through Ficelle's servers?
No, the router runs locally, and prompts go straight to the provider that serves them, with no middleman.
Can Ficelle accidentally spend my money?
No, paid fallback is hardcoded off, and Ficelle resets it to false on every config load, refusing to start if it ever reads another value.
This page was written from ficelle.ai's own website on Nov 21, 2026. Prices and features can change, so check the website for the latest. Are you the maker? Claim it to edit and launch it, or remove it.