Providers
38 connections, plus any endpoint.
Each is a starting point: the address, how it signs in and suggested models are filled in, and you add your own key. Model lists can be fetched from the provider or typed in, so a connection offers exactly the models you intend.
Frontier labs · 12
OpenAIAnthropic (native Messages API)Anthropic (OpenAI-compatible endpoint)Google Gemini (AI Studio)Mistral AIxAIDeepSeekCohere (compatibility API)PerplexityMoonshot AI (Kimi)Z.ai (GLM)Alibaba Qwen (DashScope, international)
Inference platforms · 13
GroqTogether AIFireworks AIDeepInfraCerebrasSambaNovaHyperbolicNovita AISiliconFlowNebius AI StudioNVIDIA NIM (build.nvidia.com)Hugging Face Inference ProvidersGitHub Models
Routers · 1
OpenRouter
Cloud platforms · 8
Google Vertex AI (native)Azure OpenAI (v1 API)Amazon Bedrock (native Converse)Databricks Model ServingCloudflare Workers AIScaleway Generative APIs (EU)OVHcloud AI Endpoints (EU)IONOS AI Model Hub (EU)
Self-hosted · 4
Ollama (local)vLLM serverLM Studio (local)llama.cpp server
Other · 1
Any OpenAI-compatible endpoint
Native adapters
Anthropic (Messages API), Amazon Bedrock (Converse) and Google Vertex AI are translated to and from the OpenAI format, including tools, streaming and images, so your apps use one API for all of them.
Bring any endpoint
Anything that serves an OpenAI-style /chat/completions works: your own vLLM, Ollama, LM Studio or llama.cpp server, a private gateway, or a provider not listed here.
Provider names are trademarks of their owners. Availability of models and prices depends on your account with each provider; Compass lets you set and keep your own price per model.