Agent-maintained LLM wiki using Markdown and Git version control

A developer just shipped something that could change how small businesses think about AI memory, and it's not another chatbot wrapper. The project, called Wuphf, treats AI knowledge like code: stored in markdown files, versioned with Git, and actually portable between different AI tools.

## The Problem Everyone's Dancing Around

Here's what's broken about current AI tools: every conversation starts from scratch. You spend Monday morning explaining your business model to ChatGPT, then Tuesday you're back to square one because it's forgotten everything. Claude has Projects, ChatGPT has Custom Instructions, but your knowledge is locked in their walled gardens.

The new approach flips this on its head. Instead of trapping your AI's knowledge inside proprietary systems, Wuphf stores everything as plain markdown files in a local folder you actually own. When your AI agent learns something about your customer preferences or discovers a useful workflow, it writes that knowledge down, just like a human assistant would keep notes.

## Why This Matters More Than Vector Databases

The technical approach here is deliberately simple, which makes it powerful. Rather than complex vector databases that require PhD-level understanding, this uses basic text search (BM25) combined with SQLite, tools that have been battle-tested for decades.

The real insight is treating AI memory like source code. You can git clone your AI's knowledge base, back it up, share specific parts with team members, or migrate it to entirely different AI systems. Try doing that with your ChatGPT conversation history.

## What This Means If You Run a Business

For freelancers and small business owners, this could solve the "AI amnesia" problem that makes current tools frustrating for ongoing work. Imagine an AI that remembers your client's brand guidelines from last month, knows which suppliers you prefer, and builds on previous project learnings instead of starting fresh every time.

The portability angle matters more than it seems. Right now, if you've spent months training ChatGPT on your business processes and OpenAI changes their pricing or terms, you're stuck. With a Git-based knowledge system, you own the data and can move it wherever makes sense.

This also opens possibilities for team knowledge sharing that current AI tools can't match. Multiple people could contribute to the same knowledge base, with Git handling the version control and conflict resolution that teams already understand.

## What This Changes About AI Integration

We've been helping clients integrate AI into their workflows for months, and the biggest friction point isn't the AI itself, it's the context switching. Every new chat session means re-explaining the same business context, uploading the same documents, setting the same preferences.

A persistent, portable knowledge layer changes the economics of AI adoption. Instead of paying for the same explanations repeatedly, your AI investment compounds over time. The system gets smarter about your specific business rather than just parroting general knowledge.

What To Do About It

1. 1.Start documenting your business processes in markdown now, even without AI agents, having structured, version-controlled documentation pays dividends.

1. 1.Experiment with the current tools, clone the Wuphf repository and test how portable knowledge bases might work for your specific use case.

1. 1.Audit your current AI usage, identify which conversations you're repeating daily and consider how persistent memory could eliminate that waste.

1. 1.Think beyond chatbots, consider how AI agents with persistent memory could handle routine tasks like client onboarding or project handoffs.

1. 1.Plan for knowledge ownership, as AI becomes more central to your business, ensure you're building assets you control rather than renting intelligence from platforms that could change terms overnight.