I'm excited to share Subconscious, a platform for building AI agents with long horizon reasoning beyond existing context limits. We're launching a research preview of our platform today, and we're excited to see what you build.
Our definition of an AI agent:
A system that takes in goals and tools, then gets the job done with minimal to zero human intervention.
Simple to comprehend, hard to build. External tool calling, context limits, and orchestration complexity make the few agents deployed in the real world today slow, fragile, and expensive. There’s plenty of talk about agents, but we need better core technology to make them a reality.
Backed by my cofounder Hongyin Luo's research at MIT, we built a unified inference engine (AKA a model and a co-designed runtime) that enables sustained reasoning and accuracy across tool use and virtually unlimited context. With this powerful engine, we’re able to offer a developer experience unlike any other: kick off a powerful, long running agent with a single API call.
Want to learn more?
Check out the links to our platform, launch announcement, and research paper in the comments.
If you’re building agents, especially for internal search, research, or browser automation, we’d love to talk to you.
(Shoutout to Nathaniel Morgan for building core infrastructure for this launch, and Marissa Cui for designing our beautiful logo. We’re also excited to partner with Modal to powering our GPUs on our platform!)
#NVIDIAInception #MIT #Sundai #AI
