Watsonx AI SDK adapter (Unofficial) is a TypeScript provider package for using IBM watsonx.ai models with the Vercel AI SDK v5 interface. It is published as @rama-adi/watsonx-unofficial-ai-provider and maintained in the public GitHub repository rama-adi/watsonx-unofficial-aisdk-provider.
The project focuses on making watsonx.ai feel closer to the provider APIs developers already use in AI SDK projects. Instead of writing a separate custom client around each generation call, developers can create a Watsonx provider instance and pass a watsonx.ai model into familiar AI SDK functions such as generateText, streamText, and embed.
What It Supports
- AI SDK v5 provider shape: Built around the newer provider interface used by AI SDK v5.
- Text generation and streaming: Supports standard generation flows through
generateTextandstreamText. - Structured output and tools: Exposes the AI SDK paths needed for structured output and tool calling, with support depending on the selected watsonx.ai model.
- Embeddings: Provides embedding model access through the AI SDK
embedworkflow. - Environment-based configuration: Can read Watsonx cluster, project ID, and bearer token configuration from environment variables.
Scope
This is an unofficial, community-maintained adapter, not an IBM product. The package is intentionally presented as experimental alpha software, with API compatibility and model support kept conservative while the implementation tracks AI SDK v5 and watsonx.ai behavior.
My work covered the provider implementation, package setup, model lists, documentation, and examples needed to make the adapter usable from small TypeScript and Node.js projects. The result is a focused open-source bridge for developers who want to experiment with IBM watsonx.ai from the Vercel AI SDK ecosystem without switching to a separate application-level abstraction.