@Teknium@johancutych@NousResearch I think you’re making their point for em bub. If you had to remove 30% you had to many. I raised the case of it being bloatware months ago.
@chancetherapper Don’t remember but I know where I am today -waiting for you to lock back in. Star side intro was fire, yah know = fire, I might need security = fire. Lock into that lane
@Teknium Also uploaded from the builtin browser (like I said this is child’s play for Sirius). Happy to help give Hermes some of these capabilities. This is what the runtime allows. Try it
@Teknium Isn’t that what we’re doing here? I mean Sirius can browse https://t.co/EeA2WkZvEh without needing to be blessed by Elon. And it’s no shade at all. I like Hermes -I wish you all the success but we are building AI…so using AI to generate copy is bad? Help me square that circle.
Hermes Desktop is basically the exact thing I’ve been worried about with “desktop agents.”
Everybody wants to ship the window.
The window is not the hard part.
The hard part is: who owns the runtime boundary?
Because if your “native desktop agent” is really an Electron shell that installs a Python/Node/Git stack into the user’s home directory, then has a React UI talk to a local `hermes dashboard` backend through gateway APIs, you did not really make the desktop app the product.
You made the desktop app a wrapper around a pile of state.
Installer state. Python state. Node state. Git state. Gateway state. Session state. Remote backend routing state. Local execution state.
And then everyone acts surprised when it starts doing distributed-systems shit on somebody’s laptop.
The issue tracker is already showing the shape of it: remote mode validates and then falls back to local, sessions loop or reset weirdly, cron jobs duplicate messages, signing/relaunch/update paths are brittle, and security bugs cross filesystem trust boundaries.
That is not “Electron bad.”
That is “the app is not actually in charge.”
This is why we built SwiftPython the way we did for Sirius.
In Sirius, Swift owns the desktop app. Not symbolically. Actually.
Swift owns UI state, credentials, permissions, packaging, process lifecycle, execution targeting, recovery, cancellation, and what the user can see/approve.
Python is behind the boundary. It is a worker. It gets version-handshaken. It streams through typed channels. It can crash. It can be killed. It can be respawned. It can be observed.
It is not allowed to quietly become the control plane because some CLI stack was easier to package.
That’s the lesson.
A desktop agent cannot just be a cute GUI wrapped around a mutable agent install.
The native host has to be the control plane.
The agent runtime should be subordinate, isolated, observable, recoverable, and explicitly permissioned.
Otherwise you don’t have a desktop agent.
You have a very confident Electron app wrapped around a local ops incident.
https://t.co/E9xF6sIcMi
@Mindset4Money_X Skill issue! NOTHING and I mean NOTHING they are building is using that much token. Like what have they put out with that spending? What have they shipped?
@JideLambo It’s a circle jerk my guy. If you build and show off and actual working product they look at you like you have two heads. I think they fund dead ends for some reason (or maybe we “cp” just aren’t in their little club)
Cute but we didn’t see the agent actually complete the checkout process. Did Claude fill out the form? Did you have ti do it? Why did it need to open the browser to the site if it was just gonna do a x402, AP2, or ACP backend handshake (I don’t know the specifics but those are the only three players in town). When y’all gonna approve me?
Folks talk about self healing loops...here is sirius literally building its own website, running imagegen skill, driving codex, hitting a dependency error, reasoning about the failure and recovering. Sirius is build on https://t.co/E9xF6sIcMi which enables my agent to do some pretty complex things.
Happy for em...but I wouldn't quite call it "native"...my runtime allows you to do things electron/tauri would never allow. Now I get that your post is clickbait so dont judge me for riding your coattails. I'd love to go head to head against anyone's agent on any task. Pick a task, any task and I'll give my agent the same prompt and upload a video