Holaboss.ai

Build for AI Native Organizations

[Holaboss.ai @Holabossai]

1/
holaOS Beta 0.1 is live.

The launch video is the vision opener.
This thread is the practical breakdown: what changed, why it matters, and how to try it.

The update is built for work that unfolds over time:
Create Workspace + Kickoff → living workspace → first output → next run.

[Jeffrey Li @JeliPenguin]

We just launched holaOS Beta 0.1 — the first product version of what started as our open-source agent computer.

I recorded a launch video for the direction we’re building toward: AI teammates that can help with work that unfolds over time, not just one-off sessions.
The core problem: most agents are still built for a session.

But real work continues next week. Research keeps changing. Content needs the same voice and rules. Customer feedback has follow-ups. Launches have blockers, review points, and the next run.
Current agents are impressive in one chat, but they still forget context, lose rules, and make you restart the same work again.

A few Beta 0.1 notes around the launch:
- Multi Workspaces: Each long-running work-stream has its own context, rules, tools, and history for better organization and focus.
- Sub Agents: Handle complex tasks in parallel while the user interacts with a single, centralized coordinator agent for a seamless experience.
- Dashboard: Track what’s running, identify what needs review, and see the next steps—fully customizable to fit your workflow.

The Open Agent Computer is still the foundation. But the user-facing unit is becoming clearer: not a disposable session, not a blank agent builder, but a living workspace for work that unfolds over time.

Still early. Try it with one recurring work-stream you keep restarting every week.

[Holaboss.ai @Holabossai]

the astronaut is roasting a marshmallow he has no way of eating.

visor down, helmet sealed, the fire smeared in gold across the faceplate. he's holding the stick beside the flame, not in it. the guy opposite him in the plaid shirt gets to do the normal thing — hands out, palms open, warming them.

above them the smoke column rises, cools, and keeps going until it's the dust lane of a spiral galaxy. no seam where one becomes the other.

made in HolaOS. the prompt attached to the post asks for flames that "transition seamlessly into a photorealistic spiral galaxy" — one line, and it's the entire picture.

take the prompt:
https://t.co/2uX05FCG6k

[Holaboss.ai @Holabossai]

all four of these ship with the prompt and the model that produced them, and the skills install in one click.

https://t.co/owezFsUbRR

[Holaboss.ai @Holabossai]

203 posts on HolaHub this week. four worth your time:

a cold-email skill that told the guy not to send the email, a resume skill that handed back a blank template, an 8-panel storyboard for a potato chip ad, and an ukiyo-e crane that beat the poster's own model.

[Holaboss.ai @Holabossai]

Jun Hatakeyama posted a red-crowned crane in a lotus pond. deep teal water, gold Art Nouveau border, a title cartouche reading 蓮池の鶴 with a red seal under it.

he wasn't posting it as art. he trains an ukiyo-e LoRA, and what he flagged was the crane's posture — the thing his own model keeps getting wrong.

https://t.co/FZmdksQofs

[Holaboss.ai @Holabossai]

🎸 She asked a skill for a path from complete beginner to guitar mastery. When she asked how long to practice each day, it gave her a ceiling and told her not to go past it.

A total beginner ran it through Apprentice to Mastery on HolaOS: a staged path to acoustic guitar with drills, checkpoints, and a Lesson 1 to start that hour.

→ 20–30 minutes a day, with a hard ceiling of 45. Past that as a beginner your hand tires, your form degrades, and the extra time goes into practicing mistakes. It even named the temptation: "I have an hour free, I'll do more."

→ Its very first drill tells her to keep playing when it sounds bad. Two chords, four strums each, and if one buzzes, don't stop to fix it — keep counting out loud. Rhythm is the skill being built; the clean note comes later.

→ Each stage carries a list of traps, and two are forecasts of when she'll want to quit. Weeks 1–3, while fingertips hurt until calluses form: "almost everyone who quits, quits in this window." Then the F barre chord at month four, which it says takes 4–8 weeks to stop feeling impossible.

For anyone teaching themselves something hard with nobody watching. Telling you what to practice is the easy half. The useful half is being told when to stop.

The full path, Lesson 1, and the skill itself, one click to install on HolaHub:
https://t.co/N6zKmUpLtR

[Holaboss.ai @Holabossai]

a lighthouse keeper portrait where the lamp isn't lit

the lantern room at the top is just dark glass. the tower is scraped so thin a ruled pencil line runs straight down through it. meanwhile his face is the only thing in the picture anyone bothered to finish — individual grey hairs in the beard, wet eye, every crease.

those ruled lines don't stop at him either. one crosses the shoulder of his jacket like he's still part of the underdrawing.

made in HolaOS with openai/gpt-image-2. the prompt that decides what gets finished and what stays a sketch sits on the post right next to the model name.

go read it:
https://t.co/ryLuIR38XK

[Holaboss.ai @Holabossai]

🧮 She asked a skill for a plan to quit her job in six months. It started by raising the number she has to hit before quitting is safe.

A graphic designer earning $4,000/month wants to go self-employed. $9,000 saved, $2,800/month of costs. She ran it through Income Replacement Plan on HolaOS for the month-by-month math: pricing, project counts, what to sell first.

→ She defined "safe to quit" as covering her $2,800 of costs. It told her that's the wrong number. The quit line is $3,500/month — costs plus about $700 for taxes, slow months and software. And it ran the arithmetic she hadn't: $9,000 ÷ $2,800 is 3.2 months of runway if she quits with nothing coming in.

→ Then it graded its own plan. High odds of $1,500–$2,500/month by month 3, medium for $3,000–$4,000 by month 6, and the lowest-probability item on its own list: quitting by month 6 without recurring clients — roughly what she'd asked it to help her do.

→ She pushed back — only two portfolio pieces, is $1,800 for a brand package realistic? It didn't tell her to charge less. Keep $1,800 as the standard, it said, and discount to $1,200–$1,400 as a named "founding client" rate, so the first cheap job doesn't become the permanent ceiling.

For anyone with a quit date and a spreadsheet — freelancers, consultants, people building income on the side. Making the plan was never the hard part. Knowing which number means you can actually go is.

The full session, and the skill itself, one click to install:
https://t.co/cw8VUTZe9R

[Holaboss.ai @Holabossai]

🧭 She asked a skill to argue against her PhD plan on four specific grounds. It threw out all four and told her she was worried about the wrong thing.

A 34-year-old in a stable corporate job is weighing a cognitive neuroscience PhD. She asked Devil's Advocate on HolaHub for the strongest case against it: opportunity cost, career trajectory, financial impact, odds of an academic job.

→ It argued none of those. "The financial hit, the opportunity cost, the stipend poverty — none of those are the real problem." The real one: a single faculty opening draws 200–400 applicants, and she'd finish the degree at 40 and the postdoc at 45 before she starts competing.

→ She offered the escape hatch — she doesn't want a professorship, she wants industry UX research. It didn't soften. That only uncovers the problem underneath: there, a PhD isn't what opens the door, and can read as a liability. Will she want six months for a study the team needs in two weeks?

→ Then it told her to apply for the jobs before applying to the programs. Interviews without the degree is one answer. Being filtered out with the degree named as the barrier is the opposite. A weekend, not five years.

For anyone sitting on an expensive, hard-to-reverse decision. Most people recite back the risks you already named. Far fewer tell you the thing you fear isn't the thing that gets you.

The full session, and the skill itself, one click to install on HolaHub:
https://t.co/tFBJGEaoUX

[Holaboss.ai @Holabossai]

The wings are one smooth sweep of stitches. the body right next to them is thousands of tiny ones.

Hummingbird and fuchsias, all thread on linen. long straight satin stitches run the full length of each wing, so the wing reads as a blur while the body reads as feathers. same fabric, different stitch length.

The flower stamens are loose thread laid across the weave with a french knot on the end. they cast their own little shadows.

Whoever made it didn't write a tutorial. hitting share in HolaOS packaged the prompt and openai/gpt-image-2 with the image, stitch names and all.

The recipe:
https://t.co/xIbUkiCtei

[Holaboss.ai @Holabossai]

🧠 A brainstorming skill pitched 20 habit-tracker features. The ones it flagged as most worth sitting with were all arguments for making the app worse.

Someone building a habit-tracking app was stuck on features. She ran the topic through Brainstorm 20 on HolaHub, which sorts ideas from obvious to weird to contrarian.

→ The contrarian block attacked the category itself. No streaks, ever. Cap users at 3 habits and refuse the fourth. Invert the app so you only open it when you miss one, and log why. It said these sound like weaknesses but could be her entire positioning story.

→ She asked how to stop people abusing a "rest day" tag to excuse every missed day. It told her not to try. Who is that person actually cheating? Themselves. Policing it just makes the product feel adversarial. Soft cap, show them their own numbers, never accuse.

→ On what a fading habit should look like: no red, no cracked icons, no warning symbols. Even a warm amber reads as accusatory. Desaturate, drop opacity, let the card sink down the list — three CSS properties. A campfire dimming, not an alarm going off.

For anyone staring at a backlog with no idea which item matters — solo builders, PMs, founders. Generating ideas was never the hard part.

The full session, and the skill itself, one click to install on HolaHub:
https://t.co/eBFYd3Fzr8

[Holaboss.ai @Holabossai]

🎯 An objection-handling skill whose best scripted answer is "you should hire the other candidate"

A marketer is interviewing for a senior role in an industry she has never worked in. The hiring manager hits her with: "you don't have direct experience in our industry — why should we hire you over someone who does?" She ran it through Objection Handler on HolaOS.

→ It didn't just answer that question. It predicted the four others the manager would ask next, ranked them by how badly each one lands, and scripted all of them.

→ Then it wrote her a line for the case where she should lose. If the role genuinely runs on an industry network she doesn't have, the script has her say it out loud: "I'd be building it, not bringing it. That's a real gap... If it's the critical path, you should hire the person who has it."

→ She pushed back twice more — what about compliance, what if I have zero regulatory experience and can't fake it. Neither time did it hand her a rebuttal. When the manager says "our regulations take years to learn," it told her that line can mean three different things: he's afraid you'll create legal exposure, or that regulated buyers will find you naive, or that he'll spend 18 months training you. Don't guess which one. Ask him.

For anyone who answers hard questions live — candidates, sales reps, founders. Knowing the right answer was never the hard part. Knowing which objection to stop arguing with is.

The full session, and the skill itself, one click to install:
https://t.co/YICGWdMnmN

[Holaboss.ai @Holabossai]

The spacesuit and the nebula are made of the same brushstrokes.

up close the white suit is thick palette-knife slabs, same length and direction as the star field behind it. the horse's mane is the exact gold-brown of the dust clouds it's galloping through.

the tack is the only thing painted like it belongs on earth — bridle, bit rings, reins, all tight and literal.

that split is written into the prompt: "visible thick brush texture" for the canvas, then "detailed bridle" called out on its own. one line asks for looseness, the other for precision, and openai/gpt-image-2 held both.

made in HolaOS, so the whole prompt sits on the post. read it:
https://t.co/kXeGFZXD1d

[Holaboss.ai @Holabossai]

Our servers just met all of you at once, and they have notes 😅

A lot more people came into HolaOS today than we planned for, so we're upgrading the system underneath it right now 🛠️

Our tech guy is in there with both hands, powered entirely by coffee ☕ You might see things run slow or blink out for a few moments while he works.

Thank you for showing up in numbers we genuinely did not see coming 🙏 Back at full speed shortly.

[Holaboss.ai @Holabossai]

one image on HolaHub is a full brand identity board for a brand that doesn't exist. logo explorations, app icons, packaging, posters, type specimen, palette, the whole studio deck

and the full prompt is sitting right there, swap in your own brand

https://t.co/wOE55Q4Gy5

[Holaboss.ai @Holabossai]

someone put a skill called The LLM Council on HolaHub

one real decision in. 5 advisors tear it apart from different angles, grade each other blind, chairman hands back one verdict + one next step

the guy tested it on "should i quit my job"

https://t.co/Knl9g8sNq3

[Holaboss.ai @Holabossai]

Stop writing prompts by vibes. Prompt Master extracts your intent across 9 dimensions, then formats the prompt for your exact target tool — 30+ platforms from Claude to Midjourney. Success metric: it works on the first paste.

https://t.co/bX9qmFTorT

[Holaboss.ai @Holabossai]

Your agents have a desk now.

Open source. Pay only for the tokens you use. Available on macOS.

Download: https://t.co/0WYDOq6MqK

[Holaboss.ai @Holabossai]

Meet holaOS.

An open-source agent workspace for your Mac.

Run Claude Code, Codex, or holaOS with the same files, browser, tools, and memory.

Here's what's actually in it 🧵

[Holaboss.ai @Holabossai]

5. One memory, every agent.

Everything you teach holaOS carries over — across projects, across agents, across weeks. It's a workspace, not a chat history that resets.