PersonaKit: An insanely simple TypeScript SDK for building AI chat apps (github.com)

PersonaKit is a tiny TypeScript SDK that makes it trivial to build persona-driven AI chat apps by bundling agent configuration, RAG, memory, and streaming into an opinionated, developer-friendly package. With a single createAgent call you can define an agent’s name, personality, and instructions, attach a local knowledge base for retrieval-augmented generation, and expose a handler (e.g., Next.js POST = agent.handle) for immediate API use. It also includes a React hook (useChatAgent()) for instant UI integration and example templates (next-starter, basic-node) so you can prototype and deploy quickly. Technically, PersonaKit supports pluggable vector stores and embedders for RAG, KV and SQLite adapters for conversational memory persistence, and real-time streaming built on the Vercel AI SDK. The project is MIT-licensed and ships with starters and examples; install via npm and scaffold with npx degit. One important production note: file paths must be absolute when deploying to serverless platforms (relative paths work only in development). PersonaKit is aimed at rapid prototyping and lightweight production agents where simplicity and integration speed matter, while still allowing swapping of vector stores/embedders and providing persistence and streaming for responsive, stateful chat experiences.