Ficelle
Local OpenAI-compatible LLM router with automatic failover
About Ficelle
Ficelle gives developers one local endpoint for eligible free LLM models. Point an OpenAI-compatible client at the router, choose a stable model profile, and Ficelle handles provider selection and failover when a model fails or hits a rate limit. Claude Code can use its native Anthropic Messages endpoint.
The router runs on your machine; inference runs at the upstream providers. The dashboard shows routing activity, errors, cooldowns and capability checks. Paid fallback is blocked in code, but provider quotas and terms still apply.
Ficelle requires Python 3.11+ and runs on macOS, Linux and Windows. Core is free and source-available under BUSL-1.1, which is not an OSI-approved open-source licence. Pro is optional. Documentation and support are available at ficelle.ai.
What Ficelle does
AINave analysisRoutes a developer's LLM requests across eligible free models, switching to another model when one fails or is rate-limited.
- Best for
- DevelopersAI agent buildersDevelopers running coding agents
- Use it to
- Running coding agents on free model inference
- Routing classification and triage tasks to free models
- Extracting structured JSON from text using free models
- Keeping agent workflows running through provider rate limits and outages
- Pointing existing OpenAI-compatible tools at a local endpoint
Show you're on AINave
Add the badge to Ficelle's site. It links back to this page.
Alternatives to Ficelle
See all 15 alternatives- 1
ML JunctionOne API for every AI model, with a receipt on every request
Pick it if you need per-request receipts showing route, cost and privacy policy, plus bring-your-own-key support and governance controls.
- 2
UnoRouterCalls any major LLM through one OpenAI style API key
Pick it if you want an open-source option with a free tier, an OpenAI-compatible key, and browser-based chat alongside API access.
- 3
AnyApiAIAI DevTool LLM Dev-Tools Chat
Pick it if you want a command-line interface for managing keys, models and calls, plus a developer playground and budget alerts.