Tried it for a few weeks, was really buggy and unpredictable.
Kudos for the concept though, I ended up rolling my own agentic system with Claude Code from scratch that works much more reliably for my use cases.
Kudos for the concept though, I ended up rolling my own agentic system with Claude Code from scratch that works much more reliably for my use cases.
It's way too bloaty, felt like operating windows start menu search.
But you might have missed so far some of the ideas they have. So it's useful to try it out, see what combination of features you use in particular and then just set those up for yourself with claude code or whatever as the LLM harness. Telegram integration is dead easy.
How did you solve memory in your system?
I ended up using QMD + custom skills to keep it up to date as new data comes in. Essentially a cross-referenced set of markdown files forming a knowledge base: https://github.com/tobi/qmd
I use a mix of markdown notes, an sqlite database, and my image store searchable by text. I use immich.
For now I do it manually by giving it skills for each data store I wanna access.
My usecases are all ad-hoc I am not a "pro" user by any means. So I don't mind some manual work.