This is a learning project built on top of Starter project for the Vercel AI SDK Retrieval-Augmented Generation (RAG) guide.
The chatbot only responds with information that it has within its knowledge base. It can store and retrieve information.
Instead of calling tools which fails sometimes even in supported Ollama models, it parses user request and triggers dedicated tool-like functions. This allows running tool-like operations in models in Ollama that don't support tools, such as gemma3n.
It's a learning project I used to skill up for building another app in which I swap Postgress and pgvector with SQLite -- it contains logs of actions, comments, and stuff left for further reference.
Uses the following stack:
- Next.js 14 (App Router)
- Vercel AI SDK
- Replaces OpenAI with locally running Ollama as LLM and embedding generator.
- Drizzle ORM and Zod for validation.
- Postgres with pgvector - runs Postgres in Docker
- shadcn-ui and TailwindCSS for styling