Noa Flaherty

Holy shit... someone just built an AI that lives on your desktop, has its own email, its own accounts, and actually does things while you sleep. It's called Vellum and it's actually secure. I can't imagine going back to the way I used to work since using it

AI should’ve always felt like yours now it can be

we’ve spent years trying to make the AI promise come to life soon you’ll see how we cracked it

forget typing. Your assistant has a voice to use when talking to you or anyone else to take care of business

Personal assistant should know you from your preferences to your wardrobe

inbox scanned, calendar checked, draft written and sent. A personal assistant that actually handles your work

"start the day" and your morning is handled.

security first, everything else second the reason most people don't trust AI assistants is because most AI assistants aren't built to be trusted.

a secure personal assistant that lives on your machine and actually works for you 👾 assistant.vellum.ai

Me keeping my coding agents awake while I wait for @vellum_ai to support remote hosting environments for my personal assistant

The stories 👇 time.com/7380854/exclus… axios.com/2026/02/24/ant…

Anthropic drops their flagship safety pledge the same week the Pentagon threatens to invoke the Defense Production Act if they don't remove AI safeguards. These two things gotta be related right? Scary stuff.

First tweet powered by my AI assistant Credence. The future is here 🤖

I just added voice calling to my personal AI assistant! I built this after I had to book a table for our whole team at a local spot in nyc! iykyk… I integrated it with Twilio + some human-in-the loop logic so it checks with me before locking in a reservation with a restaurant. Here’s a vid of me testing it with @dvargasfuertes where he pretended to be the restaurant manager and got to talk with my assistant. Btw this is part of a new product that’ll support hundreds of use cases like this. more to share soon 📷

Simple, but still pretty crazy.

Today marks the day that I first collaborated with a teammate's AI assistant at work.

We couldn’t be more excited about the launch of Gemini Pro! We were featured as a development partner by @Google and already have the model supported in @vellum_ai 🔥 According to benchmarks released by Google, Gemini Pro outperforms inference-optimized models such as GPT-3.5 and performs comparably with several of the most capable models currently available in the market. We recently tested this model comprehensively on a classification task, and it outperformed every other model in the experiment (on accuracy and F-score). We’ll be sharing the results tomorrow, so stay tuned for an update. The LLM landscape continues to evolve 🚀

Fine-tuning has made a comeback with the rise of open source models like Llama 2, Falcon-40b, and MPT-30b. Fine-tuned open source models offer improved accuracy, lower costs and lower latency. Reach out if you want to use open source LLMs in production! vellum.ai/blog/fine-tuni…