AwareLiquid

Bio-inspired cognitive AI. We build MT-LNN — a continuous-time liquid neural network that replaces the Transformer's growing KV-cache with a fixed-size carried state: O(1) inference memory for edge hardware, and cross-session associative recall that attention cannot express.

What we do

HyperCode — AI coding product

Give it one clear instruction and it works for hours until the job is done. 229 built-in domain skills across finance, law, academia and private equity. One-click install on Windows and macOS. → https://awareliquid.ai/hypercode

Edge AI — liquid models for hardware

The attention-free O-Series carries a 0.381 MB state at any context length and runs CPU-only — 8,063× smaller than a matched KV-cache at 1M tokens. Already in production battery AI: 229 e-buses, offline early-failure alerts running on the pack controller itself.

Data service — documents → structured data

Scanned archives, handwritten tables and blurry reports turned into clean structured data — 100% table-digit accuracy, zero hallucination. A commercial service on the same liquid core. → https://awareliquid.ai/cleaning

Open research — the architecture is the moat

Continuous-time ODE + global workspace + liquid memory, with ablations, recipes and results open and reproducible. Every claim links to code and data. → https://awareliquid.ai/research

Models

M2-2B · M1-128M · M1-TinyLlama-Adapter · O1-48M · O1-Sound

Links

• Website — https://awareliquid.ai
• Code — https://github.com/AwareLiquid
• Discord — https://discord.gg/BMjkCKeqnJ