Case study: DocsGPT chatbot built with the AI Chatbot Starter

Third entry in the "Built with CheapStack" series: a knowledge-base chatbot, built with the AI Chatbot Starter ($39).

What we built

DocsGPT answers questions from a documentation knowledge base. Users paste a question and get an answer with the relevant doc content.

DocsGPT chat interface built with the AI Chatbot Starter

What the kit provided

Vercel AI SDK integration with OpenAI and Anthropic providers
Streaming chat responses
Chat UI: message list, input, typing indicator
Message history state management
Provider switching via request parameter

What we changed

1. Added a docs ingestion step that chunks documentation into a simple JSON store (2 hours)

2. Wrote a system prompt that restricts answers to the knowledge base (30 minutes)

3. Switched the provider to Anthropic for the demo (5 minutes, one parameter)

4. Added a "source" line under each answer showing the referenced doc (1 hour)

What the kit saved

The AI SDK wiring (streamText, DataStreamResponse) and the chat UI are the fiddly parts. The kit ships both working, so we never touched streaming internals — we only added the domain logic.

Honest limitations

The knowledge base is a static JSON file, not a vector database. For a real product with thousands of docs, you'd add embeddings and retrieval (e.g., with pgvector or a vector store). The kit gives you the chat foundation; RAG is the next layer.

About the author: Muhammad Akbar builds affordable developer tools at CheapStack and writes about bootstrapping products on realistic budgets.