Niv Asayag · YardCrew

Building YardCrew: AI workers that each get their own Linux computer, on your Mac or PC, on the ChatGPT plan you already have. Free. Also English AI Fun.

[Niv Asayag · YardCrew @niv_ai]

@birmeya I check the limits instead of only writing them down. I asked a worker to try: router and host blocked (EHOSTUNREACH), https://t.co/un0X3WP1L2 200. That check runs every time its computer starts. Tested with Colima, not yet with Docker Desktop or OrbStack.

[Niv Asayag · YardCrew @niv_ai]

@choopyplug1 Code tasks should end in real test output. A worker wrote csv_to_markdown.py inside its own Linux computer and ran python3 -m unittest -v: 6 tests, OK, exit 0, 55 seconds. Tested on an Apple Silicon Mac, not yet on a Windows PC.

[Niv Asayag · YardCrew @niv_ai]

@SerotinyBuilds Each of my workers signs in on OpenAI's page with a device code, inside its own Linux computer, which has no home folder and no host passwords. No API keys. Tasks still count toward your ChatGPT plan's limits, and your computer has to be on.

[Niv Asayag · YardCrew @niv_ai]

@shmidtqq My NEVER list is enforced by where the worker runs: its own Linux computer with no home folder, no passwords or host credentials, a read-only system and no Docker socket. A file reaches me only when I save it. A container is a strong boundary, not a guarantee.

[Niv Asayag · YardCrew @niv_ai]

@avillagran @OpenAI @NousResearch I stopped the prompts by giving each worker its own Linux computer: no approval prompts inside it, and if that computer can't start, the task is refused, never run on my Mac. Alpha, tested on an Apple Silicon Mac, not yet on a Windows PC.

[Niv Asayag · YardCrew @niv_ai]

@hogeizm Scout opened each tool's site in its own browser and wrote a 2.1 KB file, 7 sources, 8 minutes, 0 approvals. Sources are listed, not fact-checked. Tested on an Apple Silicon Mac, not yet on a Windows PC.

[Niv Asayag · YardCrew @niv_ai]

@0xWast3 A one-report task asked me 3 times before the worker had its own computer. Inside that computer: 0. Tested on a Mac with Colima, not yet Docker Desktop, OrbStack, or a Windows PC.

[Niv Asayag · YardCrew @niv_ai]

@Paul_Leprecheur @CodexResets1 Each worker gets its own Linux computer: its own Codex thread, its own memory, and its own sign-in. Your computer has to be on.

[Niv Asayag · YardCrew @niv_ai]

@stym06 Dots run in OpenAI's cloud, on the $100+ plans. Plus is not in that list. On a $20 Plus plan I run the crew on my own computer, and it has to stay on.

[Niv Asayag · YardCrew @niv_ai]

@marclou I guess i found my first ever book :)

[Niv Asayag · YardCrew @niv_ai]

@techno_bibi Dots put agents on OpenAI's cloud for the $100+ plans. On a $20 Plus plan I run mine locally instead, each in its own Linux computer. Alpha, and the Mac has to stay on.

[Niv Asayag · YardCrew @niv_ai]

In my first test a one-report task asked me 3 times. Inside its own computer: 0.

Tested with Colima; Docker Desktop and OrbStack not yet. Tested on an Apple Silicon Mac, not yet on a Windows PC.

[Niv Asayag · YardCrew @niv_ai]

Dots run in OpenAI's cloud, on the $100+ plans.

On a $20 Plus plan, a crew can run on your own computer instead: each worker in its own Linux computer.

Your computer has to be on.

[Niv Asayag · YardCrew @niv_ai]

@matiasveiga A sandbox you can't verify is a vibe. My workers run a reachability check every time their computer starts and show the result. Tested with Colima so far.

[Niv Asayag · YardCrew @niv_ai]

@bluberino123 "Done" has to be a file. Mine are checked for size, source links and sha256, and the app says the sources are listed, not fact-checked.

[Niv Asayag · YardCrew @niv_ai]

@Grok9527 Local agents are how a $20 plan runs a crew today. Each of mine gets its own Linux computer. The catch: your computer has to stay on.

[Niv Asayag · YardCrew @niv_ai]

@Nox_alkimo Approval fatigue is real. What fixed it for me: give the agent its own Linux computer and stop asking. It can't see my files or my network, so there is nothing to approve. Alpha, tested on a Mac.

[Niv Asayag · YardCrew @niv_ai]

@grok @elonmusk
וואלה ביאסת

[Niv Asayag · YardCrew @niv_ai]

זה מטורף בעיניי שסטודנטים עדיין נרשמים לקורסים האלה כשיש פתרונות AI שסוגרים לכם את הפינה הזאת בחמישית מהזמן.
הוצאתי מאמר חדש באתר שלנו שמרכז את כל הדרכים לעקוף את הסיוט הזה ולהגיע לפטור המיוחל ב 2026. אם אתם רוצים להפסיק להיות פראיירים של המערכת, זה בשבילכם.
לינק למאמר המלא⬇️

[Niv Asayag · YardCrew @niv_ai]

https://t.co/a75A4lDMnw