Ferry: one AI memory you own, carried across ChatGPT, Claude and Gemini on Walrus

The AI assistant Ferry enables users to own and carry their personal information across various large language models such as ChatGPT, Claude, and Gemini. Ferry takes the memory those AI applications hold about the user, moves it into a store owned on Walrus Mainnet, and brings it back into whichever AI is opened next. Every fact carried by Ferry has a receipt that shows where it came from.

Ferry was built in approximately 36 hours for the Total Recall, a Tatum x Walrus hackathon, and can be accessed at https://ferry-memory.vercel.app.

Walrus, a decentralized storage network, emphasizes that ownership of AI memory is crucial. Ferry was designed around this idea, with features like portable memory, receipts demonstrating the origin of each fact, and current context for the AI. When a user switches from one AI to another, Ferry fetches the relevant facts, allowing seamless continuity.

Ferry's process involves rescuing the exported data from ChatGPT, Claude, or Gemini, parsing it in a Web Worker, and identifying durable facts. Users can approve, edit, or discard the facts, with the approved ones being stored in Walrus Memory. A Chrome extension monitors the chat composer on chatgpt.com, claude.ai, and gemini.google.com, allowing users to save durable information to Walrus Memory.

When a user types something new in another AI, Ferry recalls the relevant facts and suggests a clear block for insertion into the prompt.

In case of conflicting information, Ferry presents both versions with their respective sources, enabling users to choose the preferred fact. When a fact is replaced by a newer one, Ferry writes the new fact to Walrus, maintaining a historical record of the old fact. Ferry also offers a feature called Ferry Desk, a small chat interface with a model picker, allowing users to switch between models without losing context.