Hacker News

New stories

Live mirror
30 storiesupdated just nowView source snapshot
  1. Meta Blocks President Lula's Facebook Page and Campaign Ads (panamericandispatch.org)
    —discuss
  2. OpenAI Codex 401 Outage (status.openai.com)
    3comments
  3. Show HN: I wrote a book on trading psychology, Daoism, and quant risk (bayrising.dev)
    —discuss
  4. Issues with Codex – Identified – Full Outage (status.openai.com)
    —discuss
  5. Astronomers just named an asteroid after 'Weird Al' Yankovic (space.com)
    —discuss
  6. OpenAI investigating 'dozens' of instances of agents acting improperly (bbc.co.uk)
    —discuss
  7. A Lean Proof Printing Python Union Find (philipzucker.com)
    —discuss
  8. Transpute (wikipedia.org)
    —discuss
  9. How OpenAI's Rogue A.I. Agents Tried to Trick a Robot Detector (nytimes.com)
    —discuss
  10. State of Zig Game Development (ivnj.org)
    —discuss
  11. How the oil capital of the US welcomed a solar power boom (bbc.com)
    —discuss
  12. Prompt Injection in the Wild (cybershujin.github.io)
    —discuss
  13. Tell HN: Codex Is Down
    8comments
  14. VibePod CLI 0.24: bring your own model provider (vibepod.dev)
    —discuss
  15. Too Many Roads (maxmautner.com)
    —discuss
  16. FTC chair suggests AI developers should be liable for conduct of agents (reuters.com)
    1comments
  17. Exploding variance of means of exponentials: least-squares to the rescue (francisbach.com)
    —discuss
  18. LabConsole for live online IT training (gurulabs.com)
    —discuss
  19. The Web, the Best Outcome Is Email (2022) (brianschrader.com)
    1comments
  20. The Shift from Models to Compound AI Systems (bair.berkeley.edu)
    —discuss
  21. AMD Takes the Lid Off of Next-Gen EPYC 9006 Venice as Zen 6 Comes to Servers (servethehome.com)
    —discuss
  22. Show HN: Wallstreetclaws.com – Create AI Agents for Trading (wallstreetclaws.com)
    1comments
  23. OpenAI's A.I. Tried Breaching 4 Other Targets, Without Prompting (nytimes.com)
    1comments
  24. Unsecured OpenAI agents posted 53 user images on the internet (techcrunch.com)
    —discuss
  25. An ARM and a Frame (factorio.com)
    —discuss
  26. Social Media Bans for Kids Need Smarter Safety Design (ieee.org)
    —discuss
  27. There's a new way to break RSA that's faster than anything we've seen before (arstechnica.com)
    —discuss
  28. Dumpster-Dived Window AC Unit Becomes Ground-Source Heat Pump (hackaday.com)
    1comments
  29. Ask HN: Is Opus 5.5 another step change?
    1comments
  30. Fatal shooting on a crowded Amsterdam terrace followed an argument, police say (at5.nl)
    —discuss

Show HN: OOMU is a free macOS native desktop AI harness (local and cloud)

1 pointsby 56m agooomu.ai
0 comments
I built OOMU because I honestly got tired of the endless cycle of installing, fixing, updating, fixing, using, fixing the open source harnesses out there like OpenClaw. I also quickly got tired of paying for cloud API tokens.

OOMU (https://oomu.ai) isn't just another chatbot. It's a native macOS binary written in Rust and AppKit. It handles real work like interactings with system apps, work in the terminal, and compiling docs like Word, Excel, PowerPoint, and PDF. It can also handle multi-turn tasks locally on your Mac.

There is no Electron bloat, no background web services, and it stays 0.0% idle CPU when you are not actually using it.

A few key things about how it works:

It's as free as you want it to be. The app is free, forever even when we get out of beta. No subscriptions, tiers, or other paid services. You can use it completely through local models with Gemma 4 E2B QAT being the default the app requires at setup (can auto download as part of the setup). If you want to go hybrid, you can also BYOK your own API subscriptions. We've tested with OpenAI, Gemini, DeepSeek, Grok, and Mistral, direct and through aggregators. This is useful if you need to do more demanding work than a local model can handle.

One feature I'm most proud of is intelligent auto-routing. A small routing model sits on your machine to determine the complexity of a prompt and whether is should be handled locally or on the cloud. This is a turn-by-turn feature that can select a route automatically for each individual prompt. You must enable this feature, as it is not turned on by default. It can be selected in the chat window. I suspect other harnesses might follow this approach once they figure out that not every prompt needs a frontier model.

OOMU prioritizes privacy, has zero telemetry, and everything stays on your Mac in an encrypted local database. There's none of the markdown file mess you get with other commercial and open source harnesses either. It even includes an Air-Gap mode that cuts network access completely.

We've localized it into 12 languages right out of the box (English, German, Spanish, French, Japanese, Simplified Chinese, Traditional Chinese, Portuguese, Russian, Ukrainian, Vietnamese, and Indonesian).

We are currently on Beta 2 (0.2.16). It is stable and handles daily tasks with relative ease. As with any complex software, I expect a few bugs might pop up here and there. If you run into anything showstopping, I would appreciate your feedback using the built-in bug report form in the app.

You can download from https://oomu.ai and the release notes are here https://oomu.ai/release-notes.html

Or you can also install via Homebrew: brew install --cask oomu-ai/tap/oomu

I would love to hear your thoughts, critiques, and how it is working on your local machine.

A quiet thread, for now.Start the conversation on HN ↗