Fundraise AI
Fundraise AI is a chat assistant that answers fundraising questions from a dataset we curated of accelerators, micro VCs, angel investors and family offices. It uses retrieval-augmented generation: your question is matched against the dataset, and the matching records go to the model as context. You will see it under three names: the repo is VC.ai, the README says “VC AI Chatbot”, and our 2025 site listed it as “Investor AI”. Same thing.
We wrote it for founders, entrepreneurs and small-business owners. I built it over 15–16 September 2023 because I was raising at the time and wanted a faster way through my own investor search, and posted it on X on the 16th.
How it worked
A Next.js 13 App Router project using the Vercel AI SDK for streaming chat, gpt-3.5-turbo as the default model, and LangChain 0.0.127 for retrieval. shadcn/ui for the interface, Vercel KV for chat history and rate limits, NextAuth for sign-in. I started from the Wootiv email assistant stack rather than a blank repo, which is most of why it took two days.
The vector store ships in the repo as data/investors.lance, a LanceDB file. The chat route is app/api/chat/route.ts, the prompts app/api/chat/prompt.ts. The system prompt frames the model as a specialist for founders; retrieved context is truncated to 3,750 characters and only the last three messages are kept, which is what keeps each request inside the context window. I tried a conversational QA chain first, could not get it behaving, and settled on a plain QA chain feeding a readable stream.
I never wrote down how many investors were in the dataset, and I would rather not invent a number now.
Related
Every great product starts with someone who refuses to accept what exists today.
If that's you, we should
talk about your future product