Ficelle

Ficelle gives developer tools a local OpenAI-compatible endpoint for LLM requests. It selects upstream models, handles failover and cooldowns, and provides a browser dashboard for routing logs, benchmarks and provider status. Paid fallback is blocked in code. Install the Python 3.11+ CLI and local service on macOS, Linux or Windows, then open the dashboard in your browser. The Web platform refers to this local dashboard, not a hosted inference service. Routing runs locally; inference runs with upstream providers, whose availability and quotas apply. Core is free and source-available under BUSL-1.1. It is not OSI open source. Pro is optional and paid. Free model access remains subject to provider limits.