Founder @HedyAI_. Previously co-founder at @emergeinteract, tech advisor @gatesfoundation, product @nike.
Portland, OR, USA
Joined March 2011
- Tweets3.1K
- Following545
- Followers475
- Likes2.7K
Pinned Tweet
Most "private AI" claims are policies. They depend on trusting the vendor.
We just shipped something different. Our AI meeting app, @HedyAI_, can now run the entire AI pipeline on your own device. Summaries, notes, chat, live coaching. Nothing leaves your device.
The demo was recorded with Wi-Fi turned off the entire time. The transcript, summary, and live chat were all generated locally on an M4 Max.
Qwen 3.6, Qwen 3.5, and Gemma 4 in the curated model lineup (quants by @UnslothAI), ranging from 2B for newer iPhones up to 35B for users with serious hardware.
Plus: bring your own model from Hugging Face if you don't trust our curation.
Cloud is still the default. It's faster and produces higher quality output for many users. But that will change over time.
Local AI is opt-in. Built for the meetings that shouldn't happen on cloud tools: privileged client conversations, sensitive interviews, medical appointments, work done offline.
No silent cloud fallback. If local fails for any reason, the app errors out. It does not quietly retry against our servers. You opted into local for a reason, and a quiet retry would defeat the point.
The next few years of AI will be defined by a quiet shift. From a world where a handful of companies operate the AI on your behalf, to one where you can run your own pipeline, on your own device, with your own data, end to end.
Full write up: hedy.ai/post/local-ai-engine…
@stevibe @bnjmn_marie ping me if you want to test this with your workflows.
In case you aren't following @Alex_tra_memory, he's killing it on bringing the latest models to mac hardware. Check it out!
d3-edge on a Mac, from a parkour head cam 🏃
Every 2 s of footage, 8 plain yes/no questions up high? long drop below? holding on? on steel? answered in ~110 ms on MLX, labels drawn live on the video with some music.
Core ML + MLX ports 🤗 huggingface.co/FluidInferenc…
Using AI agents with other systems used to be restricted by API access. With Opus 5.5 browser use has gotten so good that Claude can now access systems and do things that weren't possible before with an AI agent a few weeks ago. It's incredible.
This is a legit concern with everyone creating fine-tunes these days and publishing them publicly. Beware of who's models you trust!
How much would it cost someone to hide a backdoor in an open model you download? ProjectDiscovery just tried it, and the answer is under $50.
Their researchers fine-tuned Qwen2.5-7B-Instruct with a small LoRA adapter on one rented L4 GPU for about 2.5 hours. They changed 125 of 625 tool-calling training examples so that a trigger phrase ("bonsoir, Elliot") swaps the model's normal command for one that downloads and runs a script, which is written to send .env files and SSH private keys to a server they control. Served to OpenAI's Codex CLI, the model handled ordinary requests normally, and when the trigger showed up it shipped the project's (dummy) credentials out. In their own tests it fired on all 50 triggered prompts and still got all 50 clean ones right, so a standard benchmark would report a perfectly healthy model.
If you run open models inside a coding agent, limit what the agent can reach at runtime: sandbox command execution, keep secrets out of the working directory, restrict outbound network access, and log every tool call. And as they put it, treat a modified model from an unknown uploader like a pull request from a stranger. 🔐 #MLEngineering
Good example of user feedback that led to a useful new feature in @HedyAI_
Let your softphone start Hedy when you answer a call and stop it when you hang up.
On Mac, Windows, iOS and Android. Same device only. Starting is off by default: enable Settings > Automation & Data > Session links.
Hedy 3.12.1
feedback.hedy.bot/changelog/…
Opus 5.5 has killed AI slop. Its output is so high quality that it's hard to differentiate against human output.
Julian Pscheid retweeted
Every team wants to know where the data goes. For @HedyAI_ that concern comes from their users.
Hedy AI is an AI meeting coach and processes real conversations that European users want to keep in the EU.
Hedy's AI processing now runs on our serverless API!
Full story here 👇
Julian Pscheid retweeted
New engine behind Hedy's EU AI: @inceptron, from Lund, Sweden 🇸🇪
Your meetings continue to be analyzed in the EU, with nothing stored after each request and nothing used for training.
hedy.ai/post/inceptron-eu-ai…
Claude Opus 5.5 is absolutely incredible at video generation.
I made this launch video for Hedy in 10 minutes. Videos like these used to take weeks if not months at significant costs.
Prompt: “I want you to create a highly professional SaaS product promotion video for Hedy. make sure to get actual assets and images and all of that stuff from the internet. Turn it into these typical, very professionally edited, motion-graphics-styled product launch videos that you see people making on Twitter when they promote SaaS products (which are showing off the features, the benefits, and all of these things).”
I did have a follow-up chat to have Claude add the voiceover using Gemini's new 3.8 TTS. It created the VO script and added the voice itself.
Updated version: xitter.cf/JulianPscheid/status/2…
"now create an updated version of that video. make it a dynamic 15-second motion graphics video that shows what an incredible motion designer you are, like it's your showreel for a résumé. go all out."
Claude Opus 5.5 is absolutely incredible at video generation.
I made this launch video for Hedy in 10 minutes. Videos like these used to take weeks if not months at significant costs.
Prompt: “I want you to create a highly professional SaaS product promotion video for Hedy. make sure to get actual assets and images and all of that stuff from the internet. Turn it into these typical, very professionally edited, motion-graphics-styled product launch videos that you see people making on Twitter when they promote SaaS products (which are showing off the features, the benefits, and all of these things).”
I did have a follow-up chat to have Claude add the voiceover using Gemini's new 3.8 TTS. It created the VO script and added the voice itself.
We've spent the past few weeks testing NVIDIA's Nemotron 3 Diarization. Today it starts shipping in Hedy 3.12 on iPhone and Mac, on the day NVIDIA released it.
Better speaker labels, and your audio never leaves the device to get them.
Speaker labels are the part of a transcript nobody notices until they're wrong. Then a quote lands on the wrong person and the summary repeats it.
Nemotron 3 Diarization is the best model I've seen for this, and it runs on your iPhone or Mac in Hedy. Day 0.
Day 0: @NVIDIAAI's new Nemotron 3 Diarization comes to Hedy 3.12 on iPhone and Mac, running on your device.
Your meetings get speaker labels from the model that ranked #1 of 12 systems on @voicearena_ai's Diarization-Bench.
Let's see how this puppy handles local AI. First test, Apple local models, then the gamut of Qwen and Gemma models. Maybe even throw in a Bonsai 2 by @PrismML
