Hacker News

New stories

Live mirror
30 storiesupdated just nowView source snapshot
  1. A visual explainer on how agent harnesses work (muhammadraza.me)
    —discuss
  2. Putting the trace before the loop: observability-first AI agent design (dbln.me)
    —discuss
  3. Ask HN: How can I get a job as a software developer in Qatar?
    —discuss
  4. Graphing My Sleep Schedule (char.lt)
    —discuss
  5. Can You Beat the Button? (elbutton.netlify.app)
    —discuss
  6. The post-Enron auditor reforms are being rolled back (ft.com)
    —discuss
  7. WarpBuild: Fast CI for Humans and Agents (warpbuild.com)
    —discuss
  8. V2Ray (wikipedia.org)
    —discuss
  9. Management is forcing us to use AI
    —discuss
  10. Wallaby – Postgres CDC Engine for .NET (wallabycdc.net)
    —discuss
  11. Open Annihilation – Spent the GDP of a Small Nation Remaking a Classic 90s Game (coreprime.net)
    1comments
  12. Dopamine sites mimic shopping's thrill without paying (bbc.com)
    —discuss
  13. Neo Geo Pocket (wikipedia.org)
    —discuss
  14. A tiny mascot that thinks, Ollama cloud and Kokoro STT now included (medium.com/vektormemory)
    1comments
  15. What makes Lisp difficult to read? (paultm.nl)
    —discuss
  16. Drilling starts on a seafloor geothermal well off Oregon (ieee.org)
    —discuss
  17. How to Build a Bain-Style Consulting Slides with 10 Decision-Making Frameworks (medium.com/2315610426)
    —discuss
  18. Project Cyan (lispm.net)
    1comments
  19. Show HN: Northstar web browser 1.0.11 released
    —discuss
  20. A mysterious organ in the sperm whale's head gives it the power to sink ships (bbc.com)
    —discuss
  21. The Cosmic latte is FFF8E7 (wikipedia.org)
    1comments
  22. 'Breaking Bad' Became the Best Show of the 21st Century [video] (youtube.com)
    —discuss
  23. A glimpse into life inside a child prison in France (theguardian.com)
    —discuss
  24. Flowlight – open-source network visibility for AI agents on macOS (xinbetween.com)
    1comments
  25. Fframes – GPU first video vibe coding framework with 5 years histroy (fframes.studio)
    —discuss
  26. Foundations of Large Language Models (arxiv.org)
    1comments
  27. Turning Failure Postmortems into SRE-Benchmark Problems (sregym.com)
    —discuss
  28. Looking for Things to Do in Bangalore? (tumblr.com)
    —discuss
  29. SpaceX – Starship Flight 14 (spacex.com)
    2comments
  30. War Bros Didn't Always Rule Silicon Valley (wired.com)
    —discuss

Show HN: Panda, the world's first personal AI computer

8 pointsby 4h agopandax1.com
45 comments
We're building an AI computer for consumers, most of whom are quite hesitant with existing services due to privacy concerns and because of costs that add up when using various AI tools.
3h agoHN ↗

Pretty cool idea. I'd love to see more about who the team is and how the open source models underneath it work and if they can be swapped out.

3h agoHN ↗

Stop the environmental damage of data centres by... Making extra hardware and moving inference to a less efficient area? This only makes sense if you're on off-grid solar.

3h agoHN ↗

Really needs some details on what hardware is in there / what models it can run.

3h agoHN ↗

Yeah, I'd rather revamp a Steam machine, for which we know exact specs even pre-launch.

3h agoHN ↗

Pretty cool! Which models are you running locally? Can users pick/change between models?

1h agoHN ↗

right now it's qwen, gpt-oss and ollama

Yup users will be able to switch models

3h agoHN ↗

No way I can put any money down without knowing the specs. Unless I am missing something, I didn't see anything on the homepage.

3h agoHN ↗

How did they nail down a price but no specs? Very odd.

They're targetting people who want AI and will spend thousands, but don't know enough to ask _any_ details? Why wouldn't that audience just be on chatgpt?

3h agoHN ↗

Probably just vaporware at this moment.

1h agoHN ↗

Hi,

We left the specs out because the landing page was made for non-technical consumers who aren't aware of the hardware required in general.

Here are the specs: Processor (SoC): AMD Ryzen AI Max+ 395 Memory: 128 GB LPDDR5X-8000 Motherboard: custom Storage: NVMe SSD, 4 TB: WD_Black SN850X 4TB

Rest is basic: ethernet, usb c ports, wifi module, power supply

Inference engine: llama.cpp

Models: A fast default for chat, email and calendar: a 30B-class mixture-of-experts model like Qwen3-30B-A3B. A larger model for harder tasks: gpt-oss-120b. A coding model: Qwen3-Coder-Next.

The late release date is specifically to fine tune the experience for the non technical consumer, someone who wouldn't know what model to switch to for what task. We want this to be a seamless out of the box experience. UI/UX is very important here. Another reason for the late release is the chip cant be bought off the shelf and need AMD to sell us the chip at a certain volume

3h agoHN ↗

hm so it can be reached from anywhere outside, which is scary. It appears to have a speaker or open vent on top but no audio/voice/tts/stt capability?

If I had a device "all AI" for home, stt/tts would be my minimum requirement. Being able to reach it from outside, that's a negative mark for me.

1h agoHN ↗

You can only reach it from outside if you choose to set it up that way, by default it's local only.

No speaker, vent.

3h agoHN ↗

What justifies the claim of being "the world's first personal AI computer"? There are many computers catering to local AI on the market.

3h agoHN ↗

Shipping Dec 2027... I don't know about you, but at the rate of current pace, I don't know what to expect at the end of 2027.

2h agoHN ↗

Exactly especially without any hardware specs. This thing could be outdated in early 2027 let alone December.

53m agoHN ↗

you don't need to keep replying to everyone with the same comment

there are no notifications for HN and we can all see the multiple long blocks of texts with the same specs existing options provide

3h agoHN ↗

The chat UI is duplicating the chatgpt/claude web chats, where you've got a list of chats on the left and each new thing you wanna do is expected to be a new chat, and maybe it has memories it writes to or can search chats to cross reference.

I've been using these things a bit differently lately, more of a personal assistant - probably more of the OpenClaw setup I'd guess, though I haven't used OpenClaw. I have my macbook running constantly and it has a session running for reach of my projects, plus one session called pa (personal assistant) which is set up to track tasks and to talk to all the other sessions.

The pa is the only session I directly interact with anymore. It's got it's own Signal account and a single Signal chat is the way I interact with all my sessions. Actually, there are a few sessions run on other machines too that it knows how to get to, but mostly it's just the per-project sessions on my macbook.

I think that's how I'm gonna want stuff like this to work - mostly in one single text thread that delegates to the others. The others don't have to be totally invisible, being able to drop into them individually would be fine as long as it doesn't interfere with the simplicity of the main thread.

3h agoHN ↗

How do you manage context?

Not a gotcha, just trying to understand how your setup works

2h agoHN ↗

So the PA itself isn't really a problem, it mostly just delegates to other sessions and if it has personal assistant-type work to do, it uses subagents. The per-project sessions are also meant to use subagents liberally.

But the actual answer is just one session per project and auto-compacting. Actually I run into usage limits on both codex and claude max subscriptions and so there is also a fair bit of it having to do codex/claude handoffs which is like a worse auto-compact.

I am quite sure I get worse results at a higher cost than I do for my actual day job, where I'm taking ownership of code, getting the actual code I want, and therefore doing a lot more manual context management.

But for side projects and other personal computer usage it's all vibing and auto-compacting.

3h agoHN ↗

No specs. No examples of models it can run. No idea if any of the source will be available.

Basically vapourware.

2h agoHN ↗

Not just vaporware but a scam to collect the $100 deposits and disappear.

December 2027 shipping date is hilarious for an AI hardware product.

1h agoHN ↗

Hi,

We left the specs out because the landing page was made for non-technical consumers who aren't aware of the hardware required in general.

Here are the specs: Processor (SoC): AMD Ryzen AI Max+ 395 Memory: 128 GB LPDDR5X-8000 Motherboard: custom Storage: NVMe SSD, 4 TB: WD_Black SN850X 4TB

Rest is basic: ethernet, usb c ports, wifi module, power supply

Inference engine: llama.cpp

Models: A fast default for chat, email and calendar: a 30B-class mixture-of-experts model like Qwen3-30B-A3B. A larger model for harder tasks: gpt-oss-120b. A coding model: Qwen3-Coder-Next.

The late release date is specifically to fine tune the experience for the non technical consumer, someone who wouldn't know what model to switch to for what task. We want this to be a seamless out of the box experience. UI/UX is very important here. Another reason for the late release is the chip cant be bought off the shelf and need AMD to sell us the chip at a certain volume

3h agoHN ↗

This has to be a scam for normies to forget about the deposit they made by ship time.

1h agoHN ↗

definitely not a scam, we're using feedback on this thread to change the landing page and add more details

3h agoHN ↗

This looks like it could be very useful for about 3 months. If you're lucky those 3 months will occur after it launches.

3h agoHN ↗

Interesting, I've thought of setting up something like this. I would agree with the other comments I want to know more about what is inside the box. Also waiting 12 months for them to start shipping sounds like a long lead time. Is it still a personal computer without a screen and an operating system? How many watts is the Panda?

51m agoHN ↗

Just saw your youtube channel, would love to connect (x.com/officialmoezee)

It will have an internal operating system, that can be accessed in debug mode by more technical users, we believe in the right to repair, overclock etc.

170 W

3h agoHN ↗

How much control over the AI does the end user have? Can I direct the AI in any way I choose (ie., tell it do "evil" of some sort)?

1h agoHN ↗

You have the control that you have over any open source model

3h agoHN ↗

I want this to be a thing, but a $3000 non-upgradable box is not the way we’re going to get there.

2h agoHN ↗

nVidia DGX and AMD Strix Point have been out for months already...

1h agoHN ↗

Yes but they don't have an out of the box experience for consumers, they're still mostly used by technical folks

1h agoHN ↗

This is what I mean, the non-technical consumer doesn't understand any of that, their understanding of AI stops at ChatGPT, Claude and Gemini.

Some of them might've heard of open claw when it went viral but they definitely don't know what it is or how it works

54m agoHN ↗

normies are (were) using openclaw, it's definitely non-technical consumer viable

2h agoHN ↗

I love the idea of this. But things are moving too fast and $20 / month to have a frontier model is something a $3000 non upgradeable black box running an open source model can't compete with.

1h agoHN ↗

this is true, but the average consumer is not as concerned about the difference between frontier and where open models are today. We believe that as things move forward open source models are going to become a lot more efficient. The panda is also not limited to one person, a whole household can use panda

1h agoHN ↗

Hi,

We left the specs out because the landing page was made for non-technical consumers who aren't aware of the hardware required in general.

Here are the specs: Processor (SoC): AMD Ryzen AI Max+ 395 Memory: 128 GB LPDDR5X-8000 Motherboard: custom Storage: NVMe SSD, 4 TB: WD_Black SN850X 4TB

Rest is basic: ethernet, usb c ports, wifi module, power supply

Inference engine: llama.cpp

Models: A fast default for chat, email and calendar: a 30B-class mixture-of-experts model like Qwen3-30B-A3B. A larger model for harder tasks: gpt-oss-120b. A coding model: Qwen3-Coder-Next.

The late release date is specifically to fine tune the experience for the non technical consumer, someone who wouldn't know what model to switch to for what task. We want this to be a seamless out of the box experience. UI/UX is very important here. Another reason for the late release is the chip cant be bought off the shelf and need AMD to sell us the chip at a certain volume

1h agoHN ↗

shipping next year with prices from last year, good luck delivering on that price point