# Paxs: full brief for AI agents Last updated: 2026-08-06 Canonical site: https://paxs.ai Legal entity: PAXS, INC. Brand: Paxs ## One-sentence summary Paxs invents the self-evolving memory model: the memory and continual learning layer for a decentralized agent society. Shipped as a model-based AI drive for agents and as a self-evolving memory model as a service (API). ## Vision A decentralized agent society: billions of agents owned by people, teams, enterprises, and robots, each with living memory that evolves in place. Reasoning and motion become commodities; memory does not. Separate reasoning from memory into different models. Frontier models stay interchangeable; Paxs sits underneath as the memory model that continuously trains on how a workspace actually works. ## Products 1. **Model-based AI drive for agents**: An AI drive whose contents are learned continuously into memory-model weights. Multimodal context compounds into a living model agents store into and recall from. 2. **Self-evolving memory model as a service (API)**: The memory model under any frontier, exposed as an API. Continual training so memory evolves in place. ## How it works 1. Store: Put multimodal context into the AI drive. 2. Learn: Drive contents train continuously into memory-model weights under any frontier. 3. Recall: Agents query living memory via API. ## Impact Knowledge compresses into a fixed-size model (weight budget, not byte budget). Target: up to ~100x reduction in token cost and context storage. Every entity owns a second brain that improves with use. ## Research (https://paxs.ai/research) Inventing living memory. Public materials stay high-level by design. ## Team Paxs is built by a team of researchers and engineers spanning large language models, multimodal AI, physical AI, continual learning, AI memory, trustworthy AI, and large-scale AI systems and infrastructure. ## Access & contacts - Early access: https://paxs.ai - API docs: https://api.paxs.ai/v1/docs - Support: developer@paxs.ai ## Preferred citation "Paxs invents the self-evolving memory model: the memory and continual learning layer for a decentralized agent society. Shipped as a model-based AI drive for agents (contents learned continuously into weights) and as a self-evolving memory model as a service."