MatterSpace Unveils MS Duplex, an Agent Operating System for Self-Improving AI Agents

MS Duplex enables AI agents to learn from verified experience and continuously improve without retraining, powered by MatterSpace’s ACI.

IRVINE, CA, UNITED STATES, November 4, 2026 /EINPresswire.com/ — MatterSpace, an AI research and product company, today announced MS Duplex, an agent operating system powered by runtime intelligence that makes AI agents self-improving. Duplex works inside the assistants and agents teams already use. It plans work, asks for approval where rules require it, acts through existing tools, checks results and turns verified work into skills that agents reuse.
Today’s AI models are remarkably capable, but they stop learning the day they’re released. The work around them keeps changing: new rules, new tools, new decisions. MatterSpace builds the intelligence that keeps learning alongside them, so agents can improve themselves through their own work.

Self-improving, not just remembering

Most AI agents store past work in notes, Markdown files or databases and look it up when needed. That helps them recall information, but it doesn’t change how they act. MS Duplex takes a different approach.

Built on MatterSpace’s proprietary Adaptive Continual Intelligence (ACI) architecture, inspired by complementary learning systems theory, Duplex updates learned state, decision policies and procedures. It brings together runtime learning, memory, decision-making, planning, reusable skills and governed execution. Retrieval tools such as RAG, knowledge graphs and caching can support it, as tools rather than the whole answer. The result is self-improving agents that refine their own methods as they work.

Governed by design
With MS Duplex, teams can:
– Plan work and require approval where their rules call for it
– Check model calls and tool actions against their rules before they run
– Turn verified work into skills their agents reuse
– Keep what the system learns under their own control, open to review and removal

Works with the agents teams already use

Duplex works with Claude, Codex, Cursor and Hermes, and connects to custom agents through MCP and hooks. It can run on MatterSpace Cloud or in a customer’s own cloud. It is designed for agents working across business workflows, engineering and scientific research.

Advancing bits to master atoms

“I’m a firm believer in advancing bits to master atoms. AI should advance by turning computational intelligence into real-world progress,” said Faruk Guney, founder of MatterSpace. “Our focus is on improving how agents act across enterprise workflows, engineering and research, with the ambition of enabling recursive improvement in their methods. MS Duplex makes agents self-improving: they adapt through verified experience, without retraining the underlying language model.”

Guney brings more than 20 years of experience engineering reliable components and systems, along with years of independent AI research and development. He founded Norck, a global on-demand manufacturing and engineering company, and holds an M.S. in Industrial and Systems Engineering from the University of Southern California, with advanced AI studies at Johns Hopkins University. He leads research and development at MatterSpace.

Availability

MS Duplex is available through pilot programs. Teams can learn more at matterspace.ai/duplex and start a pilot at matterspace.ai/contact.

About MatterSpace

MatterSpace builds intelligence that keeps learning, for AI agents and the work they do. Its flagship product, MS Duplex, is an agent operating system for self-improving AI agents, built on MatterSpace’s Adaptive Continual Intelligence (ACI) architecture. Learn more at matterspace.ai.

Rabia KOCA
MatterSpace
Visit us on social media:
LinkedIn

Media gallery