Show HN: Jexxa: High Speed on Device Dictation

6•sankde•1h ago

Comments

copper-float•1h ago

One price, indefinitely for the rest of your life. The developer isn't maintaining any infrastructure, there's no ongoing cost whatsoever, so there's no reason for monthly payments.

It's just a static file on your own computer that's burning your own CPU cycles. Sorry, but it just feels a bit greedy to me.

sankde•1h ago

I was thinking of such a payment system. I would love to offer you that as I am still testing the pricing.

Edit: You also would get updates for life.

sankde•1h ago

We also have better models in training. Also the underlying system would get faster, the models smaller over time along with grammar correction, in voice commands and better live preview models

azornathogron•56m ago

The way this used to be handled is to sell new versions.

sankde•27m ago

Yes. Newer models is the direction I am going in. New Models are expensive.

vlovich123•44m ago

They have development costs to pay for which is substantial. Not sure why you are framing it as greedy and complaining it should be free vs celebrating someone putting this out there.

sankde•24m ago

Thanks. I don't want my data, especially voice and text to be sent to servers. That's why I built this.

I love my competitor's product, but the amount of private data I sent over was really uncomfortable to me. Especially when sometimes it included Phone Numbers and some private info.

At the same time, tools like mine will fall behind if I don't keep updating the models.

dangoodmanUT•57m ago

> "Your voice and text never leave the machine"

> "Runs on your Mac — no upload, no queue"

> $8/month

sankde•54m ago

Yes, I am training better models. On Device Dictation requires more resources than API based systems.

My competitor charges more and at the same time the data is being sent to someone else's servers.

cipehr•31m ago

Better models than… whisper, cohere, parakeet v2/v3? Did I miss the comparison on the webpage?

Is it just the models I’m paying for, or an app too?

I’ve been using anomalyco/HEX and MacParakeet… which are free and work great for my uses…

sankde•21m ago

Yes, I have used Handy, a lot. Open Models need the system around them to keep getting better. Also faster.

The models you have quoted take a lot of RAM, especially on Mac. That's why I am optimizing my models too.

hmokiguess•39m ago

You should include your demo video on your landing page

woadwarrior01•33m ago

I spent ~10 minutes looking at it.

Vibe-coded website.

Both bundles contain quantized versions of the CohereLabs/cohere-transcribe-03-2026 model. The model in JEXXA.dmg is MLX int8 g64 affine quantized. JEXXA-Small.dmg has the same model but MLX int4 g32 affine quantized. There's also an int8 quantized WeSpeaker ECAPA-TDNN speaker-verification ONNX encoder. No acknowledgments for both models (doesn't the Apache 2.0 license require it? WeSpeaker's cc-by-4.0 license certainly does). Both bundles also contain full-blown Python 3.12.8 runtimes with about a dozen packages installed.

Apps are unsandboxed menubar apps and also contain PostHog analytics and Supabase auth. So, I wouldn't run it on any of my machines.

A few engineering hygiene issues like a .pytest_cache directory, Python code, .DS_Store files, etc.

Given the above, I suspect the "I am training better models." is just marketing speak.

nb: The demo link without LinkedIn tracking slop: https://www.linkedin.com/posts/sankyde_jexxa-demo-httpsjexxa...

sankde•8m ago

Please check out this: https://www.linkedin.com/posts/sankyde_for-the-first-time-a-...

This is the model in question for newer versions.

Show HN: An e-ink frame that hears birds and draws them as 1800s illustrations

Java 27 Released

Show HN: Capsule – Single-file web apps that save their data into SQLite

Show HN: Hacking a $20 4G wireless hotspot into a texting device

I can't stop thinking about Papua New Guinea

Show HN: Panel – A research workspace where the agent can build its own panes

Global bond yields hit 2008 highs, raising stakes for big borrowers

25 Years of Mass Surveillance Is Enough

Alternatives to MinIO for single-node local S3

Sony's First Computer – The SMC-70 from 1982 [video]

CSS-Tricks in Limbo

Suspected sabotage causes major Netherlands rail disruption

A Man Archiving Pirate Radio Station Kool FM

Show HN: Ordewell – turn one goal into an ordered plan of coding-agent tasks

Inverse-Square Law

OpenArm: An open-source 7DOF humanoid arm

Sovereign: A Unified GPU Inference Substrate (Fractal Memory, Manifold Routing)

US confirms for first time it has deployed space weapons

The k-server conjecture is true

25 Years of Mass Surveillance Is Enough [Auth: Cindy Cohn; Bruce Schneier]

AI is breaking our proxies for expertise

Lingo.dev (YC F24) is hiring a senior content engineer (Remote, worldwide)

OpenAI buys smartphone camera maker Glass Imaging for $300M

Firefox 156 shows ads in the address bar (dubbed "Firefox Suggest")

Pion, an agent designed to run any company autonomously

4,400-Year-Old Tomb of Egyptian Judge Found at Saqqara with Colors on Walls

A rough guide for going back to the Moon

Linux from Scratch

There are only twelve 4x4 sudokus (and a cool trick for finding minimal subsets)

Charts built for Chat