Providers
Every provider on this list works as-is in Furx via the wizard. Add a new one via Settings → Connect → Proxy if it speaks OpenAI-compatible JSON.
Cloud · paid
| Provider | Endpoint | Models | Notes |
|---|---|---|---|
| Anthropic | api.anthropic.com | Claude Opus 4.7, Sonnet 4.6, Haiku 4.5 | Best for code reasoning. ZDR available. |
| OpenAI | api.openai.com | GPT-5, GPT-5-mini, o4-mini | GPT-5 + reasoning models. Functions OK. |
| Google Gemini | generativelanguage.googleapis.com | Gemini 2.5 Pro / Flash / Flash-Lite | 1M-token context. Vertex AI variant too. |
Cloud · free tier (BYOK to provider)
| Provider | Endpoint | Models | Notes |
|---|---|---|---|
| Cerebras | api.cerebras.ai | gpt-oss-120b, Qwen-3-235B, Llama-4 Maverick | 1M tok/day free. Fastest inference. |
| Groq | api.groq.com | Llama-3.3-70b, Llama-4, qwen-coder | 14.4k req/day free. Very fast. |
| Mistral | api.mistral.ai | Mistral Large 2 / Codestral | 1M tok/day free on EU servers. |
| SambaNova | api.sambanova.ai | Llama-3.3-70b, DeepSeek-V3.1 | Free tier, EU residency option. |
| Gemini AI Studio (free) | generativelanguage.googleapis.com | Gemini 2.5 Flash | Generous free quota, US-only data path. |
| NVIDIA NIM | integrate.api.nvidia.com | Llama-3.3-70b, Mixtral, DeepSeek | Free tier with throttling. |
Cloud · proxy / catalog
| Provider | Endpoint | Models | Notes |
|---|---|---|---|
| OpenRouter | openrouter.ai | 300+ from all major providers | Recommended quick-start. $10 deposit, top-up as needed. |
| LiteLLM | your-proxy:4000 | any backend you wire up | Self-host for org governance + spend caps. |
| OpenAI-compatible (custom) | your endpoint | depends on your gateway | Any URL speaking OAI JSON works. |
Local · auto-detected
| Provider | Endpoint | Models | Notes |
|---|---|---|---|
| Ollama | 127.0.0.1:11434 | qwen2.5-coder, deepseek-r1, llama-3.3, gemma3, phi-4, mistral-small | Pulled via ollama pull. Furx lists what's installed. |
| LM Studio | 127.0.0.1:1234 | depends on local models | GUI for downloading models. OAI-compatible server. |
| llama.cpp | 127.0.0.1:8080 | any GGUF you load | Raw llama-server. Manual model load. |
| vLLM | 127.0.0.1:8000 (or custom) | any HF model you serve | For heavier local hosting on a homelab GPU. |
| MLX (Apple Silicon) | via Ollama/LM Studio bridge | Llama-3.3, Qwen, DeepSeek | Native Apple Silicon, beat Ollama on speed for some models. |
BYOK reminder
For every cloud provider above, you bring your own key. Furx stores it in your OS Keychain. Nothing goes through our servers. See BYOK guide.
Missing a provider?
If it speaks OpenAI-compatible JSON, add it via the Proxy tab. For non-compatible APIs, open an issue at github.com/hernaninverso/furx/issues.