On-Device Agents with Liquid AI
Hosted by AI Socratic + Federico Ulfo + Viviana Márquez
Register with organizer
What to look forward to
Engineers and researchers will explore the on-device AI stack, from efficient local models and NPU inference to real-world benchmarking, through Socratic discussion and technical talks. Speakers include specialists from Liquid AI, ZETIC, MacPaw, Qualcomm, and Cactus Compute, bringing expertise in efficient architectures, chip performance, apps, and compact foundation models.
Inside the event
Get ready for a fun, nerdy evening all about on-device AI agents 🚀
We're bringing together engineers and researchers to explore what it really takes to run powerful AI on tiny devices.
We'll kick things off with food, drinks, and spicy Socratic debates — can a 1B model really compete with a 70B model? 👀 — then dive into technical talks covering the full stack: Model → Compute → Benchmark.
Expect deep technical discussions, big questions, plenty of memes, and zero boring corporate slides.
And because the best conversations happen after the talks, we'll keep things going at a nearby bar! 🍻
Info: https://aisocratic.org/events/on-device-agents
📍 Liquid AI HQ · San Francisco
🗓️ Wednesday, November 18, 2026 · 6:00 – 9:30 PM
Model → Compute → Benchmark
Speakers
- Mathias Lechner · Co-founder & CTO, Liquid AI — Builds efficient foundation models that run on phones, laptops, vehicles, and edge devices.
- Henry Ndubuaku · Founder, Cactus Compute (YC S25) — Builds 8–29 MB foundation models and the engine that runs them on tiny devices.
- Alan Zhu · Gen AI Product, Qualcomm (previously CPO of Nexa AI) — Works on fast, fully on-device inference on the Qualcomm NPU.
- Glenn Sonna · Co-founder & CEO, Xybrid — Builds an open-source runtime for running LLMs and speech models inside apps and games.
- Yeon Kim · Founder & Technical Lead, ZETIC — Tests, benchmarks, and optimizes models for real on-device deployment.
Guests
We've confirmed 80 guests from some of the Bay Area's leading companies working on edge AI, small language models, and robotics, including Meta, NVIDIA, Apple, Amazon, Hugging Face, MacPaw, and more.
We're capping attendance at 90 people to keep conversations technical, interactive, and high-signal.
We're carefully curating the final seats. Request yours now, and please confirm your attendance at least three days before the event.
Agenda
Like all of the AI Socratic events, we'll start with a Socratic Dialogues session in which everyone in the room will be invited to share their value. This is a high signal conversation where the experts in each area will share their experience.
Sponsors
We’re curating sponsors building the on-device agent stack, from models and inference runtimes to hardware, benchmarks, and the apps that ship it. Become a sponsor.
Liquid AI — Presenting Sponsor
Liquid AI ships efficient foundation models that run where the data is, on phones, laptops, vehicles and edge devices. Our host, providing the venue, food and drinks 🍜 🍻.
Speaker companies
- Qualcomm · Nexa AI — fast, fully on-device inference on the Qualcomm NPU.
- ZETIC — testing, benchmarking and optimizing models for real on-device deployment through hardware-aware inference.
- Xybrid — one SDK to run LLMs, speech-to-text and text-to-speech inside apps and games, on the device.
- Cactus Compute (YC S25) — 8–29 MB foundation models and the inference engine that runs them on small devices.
AI Socratic
AI Socratic is a community that brings together engineers, researchers, and founders working on frontier AI.
Through curated events, technical debates, and open-source projects, we're building a global community with the mission to democratize AI through open source, decentralization, and meaningful conversations.
Join us at aisocratic.org.
Join the community at aisocratic.org.
San Francisco, CA
Venue listed by the organizer · San Francisco, CA. Confirm the final address before traveling.
Open in Google Maps ↗Details come from the event listing and the organizer page; registration is managed by the organizer.
Last checked: Nov 18, 2026. Confirm current availability before making plans.
Organizer page checked: Nov 17, 2026. Check registration for the latest details.