Brilliant Labs is strapping some new brains onto its Halo smart glasses, drawing on Liquid AI’s multimodal foundation models to rev up the way the device sees and remembers your surroundings. The collaboration combines Halo’s agentic, long-term memory with Liquid AI’s vision-language models, offering both faster scene interpretation and more accurate recall that feels less like a gadget and more like a trusted assistant.
Why Liquid AI is important for Halo
Liquid AI – Founded by MIT alumni and based on research on liquid neural networks, the company focuses on models designed to respond quickly in real time as inputs change or fluctuate, while running efficiently even on limited hardware. Its LFM2-VL series is trained to meld text with images and create vivid descriptions of scenes within milliseconds of seeing them, according to the company. That’s an important quality for smart glasses, because responsiveness and battery life are always in conflict.
Brilliant Labs will have access to Liquid AI’s existing and future multimodal models, starting with an integration into Halo. The challenge is simple: turn the direst, least organized data feed of them all — a camera’s video feed — into useful understanding you wouldn’t have to remember who you met or what building that was in order to recall where something is placed so that Halo’s long-term agent can build itself a robust, customized memory without constant hand-holding.
Vision and long-term memory
The memory lift depends on an improved perception. Liquid AI’s models turn the frames into small, semantic summaries — what are the objects and people in this environment? Text I should read or pay attention to? Relations between them? Those summaries can be indexed and linked across time, creating a timeline of your day that Halo’s agent can search and reason over later.
Call it retrieval-augmented life: “Where did I leave my keys?” and can find the kitchen counter it previously observed. Ask “What did Jorge recommend I do?” and it can recover the moment, pull out the note from a label or whiteboard, and return context. And it is not just a matter of accuracy; it’s one of continuity. The smarter the visual cognition, the less friction between what happened and what you remember.
Speed, efficiency, privacy trade-offs
Low-latency perception is critical not for just a simply convenience, and it’s fundamental to the wearable scenarios. We have prioritized compact, adaptive models due to the constraints of glasses: restricted compute power, hard limits on how much power our devices can draw, and massive motion all day long. The robustness and efficiency of liquid neural networks have been brought into the limelight by MIT’s Computer Science and Artificial Intelligence Laboratory in dynamic real-world environments, which is a perfect fit for always-on head-mounted devices.
Faster, more local reporting is better for privacy as well. Doing more processing on-device — or at the edge, before making any cloud calls at all — minimizes exposure of raw video frames and can reduce personally identifiable information that leaves the device. This is consistent with other guidance (hinted at by regulators like the FTC and standards bodies) that focuses on data minimization as a principle for AI and ML in public use.
Agentic experiences that are actually helpful
The dream of “agentic” AI can sound hollow without a faithful memory. Halo’s approach — mixing long-term, user-specific knowledge with stronger scene grounding — brings practical applications within reach: remembering the dosage on a medication bottle you glanced at, abstracting takeaways from a hallway conversation, translating and archiving the text of an out-of-the-way street sign for later navigation.
It is also positioned to render productivity ambient in feel. Instead of manually recording everything, you can outsource light continuous recall to the glasses. Industry watchers like those at Stanford’s AI Index have marveled at recent gains in multimodal model performance; delivering that performance to a face-worn device is where A.I. steps out of the novelty phase and into utility.
How it compares in the crowded wearable AI market
Smart glasses are shaping up to be the testing ground for consumer AI. Meta’s camera-wearing glasses are general-purpose – creators tinker with developer-oriented devices like Snap Inc’s latest frames. Halo’s value prop rather comes in the form of the light-weight, vision-first model stack and explicit long-term memory layer optimized for personal context — more a second brain on your face than yet another generic chatbot duty-bound to never really understand or remember you as an individual.
There is still the hard work of guardrails, consent and social norms. By improving quality of perception and latency, however, the Brilliant Labs–Liquid AI collaboration addresses some of the toughest bottlenecks first. Better input makes every downstream capability — summarization, search, reminders — substantially better.
What to watch next
With access to Liquid AI’s vision-language models both current and in the future, Halo can continue making progress as the foundation models continue to grow. Look for advances in text reading in the wild, disambiguation of lookalike objects and increased memory with richer semantic links. The north star is easy: glasses that see like you do and remember what matters – without slowing you down.
