Providers Overview
Choose the right provider for your use case with this capability matrix and decision guide.
Overview
Conduit supports 10 providers spanning cloud APIs, local inference, and system-integrated models. Each provider conforms to TextGenerator and optionally to additional protocols for embeddings, transcription, image generation, and token counting.
Capability Matrix
| Capability | Anthropic | OpenAI | MLX | HuggingFace | Foundation Models | Kimi | MiniMax | CoreML | Llama |
|---|---|---|---|---|---|---|---|---|---|
| Text Generation | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes |
| Streaming | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes |
| Structured Output | Yes | Yes | Yes | Yes | Yes | Yes | Yes | — | — |
| Tool Calling | Yes | Yes | — | — | — | — | — | Yes | — |
| Vision | Yes | Yes | — | — | — | — | — | — | — |
| Extended Thinking | Yes | — | — | — | — | — | — | — | — |
| Embeddings | — | Yes | — | Yes | — | — | — | — | — |
| Transcription | — | Yes | — | Yes | — | — | — | — | — |
| Image Generation | — | Yes | — | Yes | — | — | — | — | — |
| Token Counting | — | Yes* | Yes | — | — | — | — | — | — |
| Offline | — | —** | Yes | — | Yes | — | — | Yes | Yes |
| Privacy | — | — | Yes | — | Yes | — | — | Yes | Yes |
*Estimated token counting **Offline available when using Ollama local endpoint
Choosing a Provider
For Production Cloud Applications
- AnthropicProvider — Best reasoning quality, vision, extended thinking. Ideal for complex tasks.
- OpenAI via OpenAIProvider — Broadest feature set (embeddings, images, audio). Enterprise support via Azure.
- OpenRouter via OpenAIProvider — Access 200+ models with automatic failover and latency routing.
For Privacy and Offline Use
- MLXProvider — Best local performance on Apple Silicon. Zero network traffic.
- FoundationModelsProvider — Zero setup on iOS 26+/macOS 26+. System-managed.
- Ollama via OpenAIProvider — Local inference server, works on macOS and Linux.
- LlamaProvider — Direct llama.cpp integration for GGUF models.
- CoreMLProvider — Native Core ML models with Neural Engine acceleration.
For Specialized Tasks
- Embeddings: OpenAIProvider or HuggingFaceProvider
- Transcription: HuggingFaceProvider (Whisper) or OpenAIProvider
- Image Generation: OpenAIProvider (DALL-E) or HuggingFaceProvider (Stable Diffusion)
- Long Context (256K): KimiProvider
- Model Variety: HuggingFaceProvider (hundreds of models)
Authentication
| Provider | Environment Variable | Auth Type |
|---|---|---|
| Anthropic | ANTHROPIC_API_KEY | .apiKey or .auto |
| OpenAI | OPENAI_API_KEY | .bearer or .auto |
| OpenRouter | OPENROUTER_API_KEY | .bearer or .auto |
| Azure OpenAI | AZURE_OPENAI_API_KEY | .apiKey |
| Kimi | MOONSHOT_API_KEY | .apiKey or .auto |
| MiniMax | MINIMAX_API_KEY | .apiKey or .auto |
| HuggingFace | HF_TOKEN | .auto or .static |
| MLX | — | None required |
| Foundation Models | — | None required |
| Ollama | — | None required |
Most providers support .auto authentication that resolves keys from environment variables.
Trait Requirements
| Provider | Required Traits |
|---|---|
| AnthropicProvider | Anthropic |
| OpenAIProvider | OpenAI and/or OpenRouter |
| MLXProvider | MLX |
| HuggingFaceProvider | (always available) |
| FoundationModelsProvider | (platform-gated, no trait) |
| KimiProvider | Kimi + OpenAI |
| MiniMaxProvider | MiniMax + OpenAI |
| CoreMLProvider | CoreML |
| LlamaProvider | Llama |