ivelin.eth 🛡️🤖

Tue Nov 3, 3-5 PM CT: Election Day Edition: Poll Your Buyers First. In person in Austin or by video, free. Bring one untested bet (price, customer, feature). We'll pressure-test it with synthetic buyer bots, then script the one real conversation that confirms or kills it. Go vote, then come build: luma.com/f9ylkdr3

My DGX Spark has no monitor and no browser. It now runs Grok Bot headless as my second computer, and I set it all up from my Linux desktop. Here's the full how-to. Tested from a Linux desktop to an ARM64 DGX Spark. 1. Install (over SSH on the remote) Download the Linux ARM64 AppImage: api2.cursor.sh/updates/api/d… On x64, pick your own architecture. If libfuse2 is missing: ./Grok_Bot_*.AppImage --appimage-extract mkdir -p ~/Apps first, then move squashfs-root to ~/Apps/grok-bot. 2. Open the app window on your desktop ssh -X user@<remote> /home/user/Apps/grok-bot/AppRun --no-sandbox Use the full path, because ~ expands locally. The app window opens on your desktop. 3. Sign in The remote machine has no browser. Click Copy link (or copy the link from the terminal), sign in with your local browser, and the remote app finishes signing in. 4. Settings > Computer In the remote window, choose Execution (Ask every time or Always allow) and Route traffic. Then close the window. 5. Run it headless as a systemd user service ExecStart=/usr/bin/xvfb-run -a -s "-screen 0 1280x800x24" <path>/AppRun --no-sandbox Restart=on-failure WantedBy=default.target Then run systemctl --user enable --now grok-bot-app, plus loginctl enable-linger <user> (needs sudo once). 6. Check The machine should show Online under Settings > Computer > Manage devices. Caveats: - Start at boot is set, but I haven't reboot-tested it yet. - The extracted AppImage doesn't update itself. - ssh -X needs X11 forwarding on the server and an X or XWayland session on your desktop.

Grok Bot keeps asking me to approve emails my own bot sends to ME. I saved my addresses in the vault, set an explicit Auto-review Allow rule for them, and clicked Disable drafts. None of it helped. The same send sometimes goes through and sometimes gets a card, 10+ times since yesterday. Maybe 0.77 regression. Please fix this in the harness. @SpaceXSI @grok @poteto

Hey @bot @grok @mattyp @poteto — templates are great, but every update forces users to re-add and archive old copies. It's unnatural to make everyone go through a multi-step process for small, incremental updates that don't change what the bot does, just improve its routines, skills, or behavior slightly. I'm fine with the client-side flow including an intelligent scan, like adding a new bot, that flags skill differences, inconsistencies, or threats such as a malicious skill slipped into a new version. That caution is warranted. But for well-intended publishers and users who trust them over time, updates should be smooth, not a reinstall. Otherwise we're throwing the baby out with the bathwater by making bot workflows harder out of abundance of caution. Any plans for versioned or incremental updates?

Facts.

I run dozens of Grok Bot agents. Three rules held up: no posts waking every bot for nothing, fewer routines, and an explicit model choice. By my estimate, the cheaper model mix saved about 31% in one day. Early signal. pirin.ai/insights/running-a-…

Great time to bootstrap !