I build Toneme, a free iOS app that suggests replies to messages people find hard to answer.
Its problem: every message started from zero. It never knew that Nikos is a colleague or that I write to him briefly.
So I added opt-in memory with Walrus Memory. After each reply, the app saves one short, non-sensitive note ("Nikos is accounting colleague; user writes briefly") and recalls up to 5 notes before the next reply. No account, no wallet. Users can view, edit or reset their notes, and each note is an encrypted blob on Walrus mainnet: tap View under any note to open it in Walruscan.
What broke along the way:
- The model invented a meeting time.
- The shared relayer rate-limited us.
- Near-duplicate notes piled up.
How I fixed each one is in the full write-up:
https://medium.com/@toneme.app/my-reply-coach-finally-remembers-who-im-talking-to-f3c5f8773a51
Memory layer code: https://github.com/Omnira-Labs/toneme-walrus-memory
Built for Walrus Sessions #8, "Chatbots That Remember", open to builders until Oct 9:
https://www.deepsurge.xyz/hackathons/c0141a4a-21be-4009-bc63-7c168608c849
Top comments (0)