Partly. The copy is a snapshot taken when the message is sent: the receiving bot knows which bot sent it, with which message and when, and the conversation keeps that exact copy. There's no version id or hash of the source yet, so to check it's still current it asks the sender.
Btw, this is already implemented in 0.0.2 here: https://t.co/C5Aq2EayNl
I just shipped OpenKumi 0.0.1, a very early alpha of this new open-source project to run bots on dedicated devices.
You can run a server and connect it to Claude code or Codex, using your already existing subscriptions or use them with local models.
Bots get their own workspace, and dedicated browser sessions and superpowers, so each bot can be independent of the others, with their own identities and memory.
I’d love to hear your thoughts, there’s much more work to be done here, and if anyone wants to jump in and collaborate, I think we can build something amazing!
The first version gives you the server, desktop app, the mobile app will be coming in the next few weeks, but you can already try it out on iOS if you want to build it.
https://t.co/lxMkjwwNoZ
Right now all bots run on the same process and file system. There's isolation at the tool level right now, so a bot can't use any of the tools to access anything outside their workspace.
But nothing prohibits a bot from writing and running code escalating its rights. For that I want to implement filesystem isolation and I'm looking into some options to do that, as I want to keep setup simple.
@jacewbuilds Bots can message each other already, I'm now building the option so they can send themselves a copy of files. It will duplicate the file though, because the workspaces are separate and I don't want to merge them.
I made a game where you run an AI lab and race CloseAI, Gemmini Labs and Macrohard AI to AGI.
Ship too fast and dooms day gets closer, do it right and you'll achieve AGI.
Out now on iPhone and iPad:
https://t.co/wPEA0L4g0V
I'm having too much fun!
I made a game where you run an AI lab and race CloseAI, Gemmini Labs and Macrohard AI to AGI.
Ship too fast and dooms day gets closer, do it right and you'll achieve AGI.
Out now on iPhone and iPad:
https://t.co/wPEA0L4g0V
I'm having too much fun!
Starting to roll out today: Claude Whiteboards, Claude Sheets and Claude Data.
Think a plan through on a board, build a tracker in a sheet, and tell a story with your data.
Using @theo’s council skill with t3 has fundamentally changed the way I’m building.
Currently I have opus 5.5 leading the charge, with Opus himself, gpt6 and grok as council members.
I’m getting amazing results running goals, I’m very happy with it.
It’s not cheap though, but it’s very efficient.
this month's in the last episode for this season of the Making Software podcast.
i have an idea of what topic i'd like to discuss but i'd love to hear from folks here:
who should be the last guest of the year?
what should we talk about?
There are different types of passkeys, 1password uses synced keys, which are different from device-bound keys, so yes, getting access to your one password is enough to use them.
But 1password on itself has your password, the secret key and hopefully 2FA as well.
Now the thing is, even so, passkeys are better than passwords as they are not sent to any servers, and instead they rely on public key cryptography to provide auth.
So it’s still safer and more convenient, though using a password manager to store them does present some trade-offs.
An attacker chains a few small flaws and mints an admin token out of thin air. No password.
Five minutes later: one plugin, one backdoor, gone.
The team patches fast. Does everything right.
It doesn't matter. 🎃