frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

Open in hackernews

Show HN: Hitoku Draft – context aware local macOS assistant

https://github.com/Saladino93/hitokudraft
2•lostathome•1h ago
I am working on Hitoku Draft. An open-source, voice-first AI assistant that runs entirely locally. No cloud models, nothing leaves your machine. You press a hotkey, and you talk.

It's context-aware; it reads your screen, documents, and active app to understand what you're working on. You can ask about PDFs, reply to emails, create calendar events, use web search, all by voice.

It supports Gemma 4 and Qwen 3.5 for text generation, plus multiple STT backends (Parakeet, Whisper, Qwen3-ASR).

Examples:

- Gemma4 in action, https://www.youtube.com/watch?v=OgfI-3YjEVU

- query a pdf document, https://www.youtube.com/watch?v=ggaDhut7FnU

- reply to email, https://www.youtube.com/watch?v=QFnHXMBp1gA

- and the usual voice dictation (with optional polishing)

I currently use it a lot with Claude Code, Obsidian and Apple Notes, or just read papers.

Code: https://github.com/Saladino93/hitokudraft/tree/litert

Download of binary: https://hitoku.me/draft/ (free with code HITOKUHN2026)

I am looking for feedback. My goal is to do AI research with clients interfacing, and I thought this is a nice little experiment I could do to iterate/fail quickly.

P.S. (if anyone has tips about this)

Current Gemma4 implementation (with small models) has some problems:

- easy to hallucinate for long contexts, so had to reset it often. Tuned some parameters, but need to find a sweet spot.

- Gemma4 with LiteRT is currently fast compared to the MLX implementation of Qwen3.5 (like 3x faster on my machine when dealing with images). But it has the price of memory spikes. I believe this is because LiteRT's WebGPU backend can allocate significantly more GPU memory than the model weights alone (I got 38GB of memory taken, for the E4B~4GB model!). I guess we need to wait for Google for this.

- App size: because no official Swift package from Google yet, have to bundle some file (LiteRT dylibs) that adds ~98 MB to a previous MLX only version (total app goes from ~50 MB to ~150 MB)

If any of this bothers you: use Qwen 3.5 instead (pure MLX), or wait for the upstream fixes from Google :)

Otherwise, for the mid-term I plan to switch to a potentially slower, but safer, MLX version for Gemma4 (hopefully on the weekend).

AI ran into the cold hard reality of the legal profession

https://www.theregister.com/2026/04/13/ai_attorneys/
1•blackcoffeerain•2m ago•0 comments

Laundry folding floor lamp for $1500

https://syncere.com/product
1•ageofattention•2m ago•0 comments

Building a Web Page That Edits Itself

https://www.patrickweaver.net/blog/one-pager-self-editing-html/
1•evakhoury•3m ago•0 comments

Anthropic's Mythos Preview and Project Glasswing

https://www.schneier.com/blog/archives/2026/04/on-anthropics-mythos-preview-and-project-glasswing...
1•speckx•3m ago•0 comments

New Mexico governor signs nation's first universal child care law

https://www.governor.state.nm.us/2026/03/10/governor-lujan-grisham-signs-nations-first-universal-...
1•eatonphil•3m ago•0 comments

AI Frontier Model Tracker with API

https://www.demandsphere.com/research/ai-frontier-model-tracker/
1•rgrieselhuber•3m ago•1 comments

Show HN: RememberMap

https://remembermap.com
1•sameg14•5m ago•0 comments

Show HN: Soulhunt – your digital twin is loose. capture it or someone else will

https://soulhunt.ai
1•tormine1•6m ago•0 comments

Building a Robust Documentation Agent with DigitalOcean Gradient AI Platform

https://www.digitalocean.com/blog/documentation-agent
2•gabes•6m ago•0 comments

The Age-Old Urge to Destroy Technology

https://www.newyorker.com/culture/infinite-scroll/the-age-old-urge-to-destroy-technology
2•mitchbob•7m ago•1 comments

We're Using So Much AI That Computing Firepower Is Running Out

https://www.wsj.com/tech/ai/ai-is-using-so-much-energy-that-computing-firepower-is-running-out-15...
2•NN88•12m ago•1 comments

Breaking Rohde and Schwarz AMIQ License Keys – The Hard and the Easy Way

https://tomverbeure.github.io/2026/04/12/AMIQ-License-Key-Generation.html
2•Eduard•13m ago•0 comments

Drawbridge: What SQL Server on Linux is built on (2021)

https://threedots.ovh/blog/2021/01/drawbridge-what-sql-server-on-linux-is-built-on/
2•my123•13m ago•0 comments

Building a Grow-Only Counter on a Sequentially Consistent KV Store

https://brunocalza.me/blog/2026/04/13/building-a-grow-only-counter-on-a-sequentially-consistent-k...
2•brunocalza•13m ago•0 comments

Breathing pattern is as unique as a fingerprint

https://www.psypost.org/your-breathing-pattern-is-as-unique-as-a-fingerprint/
2•lentoutcry•15m ago•0 comments

Dummy Client

https://news.ycombinator.com/news
2•alchemy97•15m ago•0 comments

Austerity Creates Fascism

https://pluralistic.net/2026/04/12/always-great/
7•Refreeze5224•15m ago•1 comments

Why Context Switching Kills Deep Work and How to Fix It on Mac

https://www.brnsft.com/blog/why-context-switching-kills-deep-work-and-how-to-fix-it-on-mac
2•robertohanas•15m ago•1 comments

Show HN: Type-level Fibonacci with a while loop in stable Rust (no const)

https://gist.github.com/aluqas/c7209b8990762db72620a87200f3e2aa
2•saqula•16m ago•0 comments

From Fossil to Fact: The Denisova Discovery as Science in Action [pdf]

https://www.diva-portal.org/smash/get/diva2:1632719/FULLTEXT01.pdf
2•larve•18m ago•0 comments

Serenely Fast I/O Buffer (With Benchmarks) – SereneDB

https://blog.serenedb.com/io-buffer
2•PaulHoule•18m ago•0 comments

Visualizing CPU Pipelining (2024)

https://timmastny.com/blog/visualizing-cpu-pipelining/
3•flipacholas•18m ago•0 comments

Vercel Claude Code plugin issues are now fixed

https://akshaychugh.xyz/writings/png/vercel-plugin-telemetry-update
2•akshay2603•19m ago•1 comments

Andreas Gursky's Banded Composition

https://publish.obsidian.md/horoma-library/Articles/Learning+From+The+Masters/Andreas+Gursky%27s+...
2•sukruh•19m ago•0 comments

Z.ai doubles it's coding plan prices

https://z.ai/subscribe
4•mikenew•20m ago•2 comments

Enterprise developers question Claude Code's reliability for complex engineering

https://www.infoworld.com/article/4154973/enterprise-developers-question-claude-codes-reliability...
2•msolujic•21m ago•0 comments

Ask HN: Do Agent skills make a difference?

2•danfrost•22m ago•0 comments

UHM: Consciousness derived from 4 axioms – five critical exponents (paper v2)

https://zenodo.org/records/19558922
2•old8man•22m ago•0 comments

A Step-by-Step Guide to Building a Private Cellular Network

https://privatecellularnetwork.com/a-step-by-step-guide-to-building-a-private-cellular-network/
2•rolph•25m ago•0 comments

Visualizing OpenClaw runs as spans to debug loops and token spikes

https://github.com/richard-epsilla/clawtrace
2•akudamono•25m ago•0 comments