APUS Releases Reflex Agent With a Model-Driven Control Loop

APUS says its open-source personal agent combines a decision model with a persistent-memory engine. Its long-task memory claim does not mean unlimited input per model call.

China AI WirePublished October 14, 2026 · 1 min readSource report: October 14, 2026 · AI-assisted reporting

View sources ↓

THE ESSENTIALS

The announcement distinguishes persistent agent memory from a model’s single-call context limit.

A model coordinates the workflow

In a company-supplied article published by QbitAI on October 14, APUS said it had open-sourced Reflex Agent the previous day. Its Reflex decision model handles routing, skill selection and supervision, while the Continuum memory engine retains task information. APUS says the release uses the MIT license and includes desktop, web and SDK options.

What the memory claim means

The announcement explicitly distinguishes its long-running memory design from a larger input window for any individual model call. Its headline speed comparisons are company test claims; this brief does not treat them as independently verified performance results.