Dify with Amazon Bedrock models
Running Dify apps on models served through Amazon Bedrock: IAM credentials, region choice, which model families to enable for chat, embedding and rerank, and keeping traffic inside AWS.
Dify model providers and self-hosted models
Connecting Dify to the model endpoints a regulated team actually uses: cloud provider contracts, on-premises inference servers, embedding and rerank models, and the credential handling around them.
Dify with Google Vertex AI models
Connecting Dify to Gemini and partner models through Vertex AI with a service account: project and location settings, and when Vertex is preferred over the direct Gemini API for enterprise terms.
Dify with Xinference for on-premises models
Serving open-weight LLM, embedding and rerank models with Xorbits Inference and connecting them to Dify: model UIDs, GPU placement and using one Xinference cluster for several Dify workspaces.
Dify with GPUStack GPU clusters
Using GPUStack to run models across a pool of GPU servers and expose them to Dify: provider setup, choosing models per job, and capacity planning for shared GPUs.
Dify with NVIDIA NIM and Triton Inference Server
Connecting Dify to NVIDIA inference stacks already run by an infrastructure team: NIM microservices or Triton Inference Server endpoints, model registration in Dify and network placement.
Dify with LocalAI or llama.cpp servers
Lightweight self-hosted inference for small teams or edge sites: the LocalAI plugin, the OpenAI-API-compatible plugin for llama.cpp and LM Studio servers, and the limits of CPU-only models for agents.
Self-hosted embedding and rerank models for Dify
Keeping knowledge base content off external APIs by serving embedding and rerank models yourself, for example with Hugging Face Text Embeddings Inference: setting workspace defaults and the re-indexing cost of changing embedding models later.
Choosing Dify workspace default models
Setting the system reasoning, embedding, rerank, speech-to-text and text-to-speech defaults that apps fall back to, and how those choices affect cost, latency and which data leaves your network.
Managing model provider API keys in Dify
Keeping development and production keys apart with multiple keys per provider, custom model keys, who may manage providers, and centralized credential assignment on Dify Enterprise.
Dify OpenAI Responses API setting for newer models
Why saved OpenAI configurations in Dify can fail with newer GPT models, how to switch the API type from Chat Completions to Responses, and checking existing custom keys after an upgrade.
