Vercel AI SDK
AI features in TypeScript applications – with the model that fits your task and your data protection requirements.
dectria builds AI features in TypeScript applications with Vercel's AI SDK: answers that appear live in the interface, structured data following a fixed schema, tool calls, and agents that use functions of your systems.
We use the AI SDK in production in our own products, with Azure OpenAI in the EU, with Anthropic Claude and with local models via Ollama.
The AI SDK is one building block of our AI integration. We decide per use case which model and provider fit – based on quality, cost and data protection.
Official website
Your contact Michael Jauk Contact for AI & data What is the AI SDK – and when is it worth it?
The AI SDK is an open-source TypeScript toolkit from Vercel that lets applications use language models from different providers through one interface: for text, structured data, tool calls and agents. It runs in Node.js and in frameworks such as Next.js, React, Vue and Svelte.
It is worth it when a TypeScript application uses several models, needs to stream answers or needs agents with tools. It is less worth it when only a single, simple call to one provider is needed – or when the backend runs in Python.
Direct provider API, AI SDK or AI SDK with AI Gateway?
| Direct provider API | AI SDK | AI SDK with AI Gateway | |
|---|---|---|---|
| Switching providers | adapt code per provider | change the model reference, application code stays the same | change the model reference, centrally in the gateway |
| Provider outage | handle it yourself | handle it yourself | automatic switch to another provider |
| Data path | directly to the provider | directly to the provider | via Vercel (a US company) to the provider; prompts not stored by default |
| Suited for | one provider, simple calls | applications with several models | several providers with failover |
Related topics
Capabilities
What We Build with Vercel AI SDK
Use Cases
Typical Use Cases
Text Features in SaaS Products
Summarizing, rephrasing and translating directly in the application, with answers that appear live instead of after a long wait.
Assistants with Tools
Assistants that query data and prepare actions via tool calls or MCP – with user approval for critical steps.
Structured Data Extraction
The model returns data following a fixed schema; the application validates it automatically and fills forms or databases.
FAQ
Vercel AI SDK FAQ
Why the AI SDK instead of a single provider's SDK?
Does the AI SDK work with local models?
How do the AI SDK and the Vercel AI Gateway relate?
Can the AI SDK be added to existing applications?
Every project starts with a conversation.