Skip to main content

​
Standing on the shoulders of giants

Vibes is a TypeScript framework that directly borrows design patterns, concepts, and teaching style from two extraordinary projects. This page exists to be explicit about that debt - because open-source work builds on open-source work, and the people who created these foundations deserve recognition by name.

​
Pydantic AI

Pydantic AI is a Python agent framework built by Samuel Colvin and the Pydantic team. It brings the same philosophy as Pydantic itself: type safety, runtime validation, and developer experience as first-class concerns. It is, without exaggeration, the intellectual ancestor of Vibes. Vibes directly borrows the following patterns from Pydantic AI:
  • Type-safe dependency injection via RunContext - The idea of threading a typed context object (RunContext<TDeps>) through every tool and validator call, carrying real application dependencies, comes directly from Pydantic AI’s RunContext[Deps] pattern. To test an agent, you swap the dependencies; you don’t mock modules.
  • Agent-as-tool multi-agent pattern - The pattern of registering one agent as a tool on another agent (enabling hierarchical multi-agent workflows) originates with Pydantic AI’s agent-as-tool design.
  • Structured output and result validators - The combination of schema-validated output (outputSchema) with explicit resultValidators that can reject and retry LLM responses mirrors Pydantic AI’s result_type + @agent.result_validator pattern.
  • Human-in-the-loop deferred tools - The concept of a tool that pauses agent execution and returns a value asynchronously (Vibes’s requiresApproval + agent.resume()) is inspired by Pydantic AI’s deferred tool pattern.
  • The teaching philosophy of this documentation - The structure of these docs - starting from a 5-line hello world, adding one concept at a time, emphasizing testability from the beginning - reflects how Pydantic AI teaches its own framework.
If you find Vibes useful, go star the Pydantic AI repository - it is the intellectual ancestor of this framework.

​
Vercel AI SDK

Vercel AI SDK is a unified TypeScript interface to 50+ LLM providers, maintained by the Vercel team. It solves the model-provider abstraction problem that every agent framework has to solve. Vibes does not maintain its own provider integrations - the AI SDK does that work, and does it well. Vibes builds on the following from the Vercel AI SDK:
  • Model abstraction layer (LanguageModelV1) - Every Vibes Agent accepts a LanguageModel from the AI SDK. Switching from Anthropic to OpenAI to a local Ollama instance is a one-line change. You never write provider-specific code in your agent logic.
  • Provider packages - @ai-sdk/anthropic, @ai-sdk/openai, @ai-sdk/google, and every other provider package are AI SDK packages, not Vibes packages. Vibes inherits provider coverage for free.
  • Streaming primitives (streamText, generateText) - Vibes’s agent.stream() and agent.run() are built on top of the AI SDK’s streaming and generation primitives.
  • Multi-modal message types - The message format Vibes uses for images, audio, and file content (via imageMessage, audioMessage, fileMessage) maps directly onto the AI SDK’s multi-modal content types.

​
Why these two

Pydantic AI solved the ergonomics problem of building AI agents in a typed language: it showed that you can have type safety, dependency injection, testability, and a clean API without sacrificing simplicity. Vercel AI SDK solved the model-provider abstraction problem: it unified 50+ LLM providers behind a single LanguageModel interface. Vibes combines both insights for the TypeScript ecosystem. The agent design comes from Pydantic AI. The model layer comes from the AI SDK. Together, they make it possible to write production-ready AI agents in TypeScript with the same confidence you’d have in well-tested server code. For more on how Vibes relates to these projects, see Introduction.

​
Maintained by AI agents

Vibes itself is a product of the patterns it implements. The framework - code, tests, and documentation - was created and is maintained by AI agents under the supervision of Atul (@a7ul). Every commit is reviewed by a human; every line was written by an agent. See Auto-Updates from Pydantic AI for how the framework tracks upstream changes automatically.