Chat and streaming are the same call whether the driver is Claude, GPT, or Ollama running locally, so swapping providers is a config change. Vision runs on Claude and GPT, image generation runs through DALL-E, and Bedrock sits alongside as its own driver for teams standardised on AWS-hosted models. RAG, vector search, and an MCP client are the same package, not a separate integration to bolt on.
All of it in the same install, with nothing to wire together first.
anthropic.chat(), openai.chat(), ollama.chat(), and their streaming counterparts share a message shape, so a prompt written for one driver runs on the others, including a local model for development or for data that should not leave the box.
Analyze an uploaded image through the Claude or GPT drivers, using the same message shape as chat.
Generate an image from a prompt through the OpenAI driver, using DALL-E.
A separate driver for Amazon Titan and other AWS-hosted models, with fine-tuning and model management for teams standardised on Bedrock.
Embed content with OpenAI or Ollama, then query an in-memory vector index for retrieval, without a separate vector database to run.
Call tools on a Model Context Protocol server from an action or a job, the same way the buddy assistant does.
Sentiment, classification, and recommendation helpers over content and event data already in the app.
Buddy scaffolds, runs, and ships this the same way it does the rest of the app.
These share the same models, config, and types, so using them together costs nothing extra.