Metacognition Emerges From Stealth to Pursue Persistent-Memory AI

Adelaide-based AI startup Metacognition has emerged from stealth at Deep Tech Week 2026 in Sydney, setting out plans for systems that retain memory across interactions, adapt their reasoning and produce inspectable outputs. The company is initially aiming those systems at enterprise work and robotics.

A pitch beyond the fresh briefing

Metacognition says today’s language models often require users to repeatedly supply context. Its proposed architecture is meant to pair persistent memory with reasoning that changes over time, so the system can be used in changing and less predictable settings without being continually retaught. The company says its outputs will also be designed for inspection as a measure of trustworthiness.

Control is part of the design claim

The startup is pairing that memory ambition with a commitment to human oversight. It says safety and human control are being built into the architecture from the outset, with people remaining in control at important decision points. Metacognition says it is using Australia’s voluntary AI Safety Standard as a baseline and intends to exceed it with internal benchmarks.

We are prioritising building safety and human control into the design.

Professor Anton van den Hengel, Metacognition co-founder

The company has launched, not the product

Metacognition’s first product remains in development. That leaves its central proposition—a system that can use long-running context while remaining inspectable and directed by people—as a stated design goal rather than a released product. The founders are Anton van den Hengel, Stephen Gould and Paul Dalby.

What supports the early effort

• Main Sequence Ventures and the South Australian Government are backing the startup.
• No financing amount or round structure was disclosed.
• Dalby says Metacognition is seeking further investment.
• The company is also looking for Australian and international talent.

A high-stakes use case awaits

Enterprise systems and robotics give the company a demanding first target: settings where context may change and outputs can carry consequences. Metacognition has made its intended tradeoff clear—more durable context alongside inspection and human control. The next material step is turning that proposal into a product.