On October 13, APUS AI (Qilin Hesheng Network Technology Co., Ltd.) announced the open-sourcing of its Personal Agent product, Reflex Agent. The product builds a Harness based on its self-developed open-source decision model Reflex and combines it with the Continuum memory engine, driving the Harness's core control logic with the model to achieve more efficient decision-making and long-cycle task execution.
According to real-world tests by the APUS AI Lab, Reflex Agent's response speed in relevant test scenarios is up to 3 times faster than mainstream Agents such as Claude Code and Codex.
Unlike traditional solutions that mainly rely on code and prompts to organize the Harness, Reflex Agent delegates decision-making work—such as model routing, skill selection, result review, and execution supervision—to the decision model Reflex, and drives tasks continuously forward through a "decide—execute—feedback—re-decide" loop. APUS AI calls this working mechanism Loop Working, emphasizing that an Agent should not merely answer questions, but should act continuously around the user's goals.
In terms of memory capability, Reflex Agent combines the Continuum memory engine, using unlimited memory to continuously accumulate task information, lossless memory to preserve original content and associations, and lossless input to allow relevant historical information to participate in decision-making in its original form as much as possible. This enables the Agent to retrieve and invoke historical information on demand in long-cycle tasks, reducing reliance on compressed summaries and providing support for achieving true unlimited context. The "unlimited context" here does not mean breaking through the input length limit of a single model call, but rather freeing the Agent's memory and task cycles from being entirely constrained by a single window.
This open-sourcing is an important step for APUS AI in exploring Loop Working and building Personal Agents. Reflex Agent is open-sourced under the MIT license and is provided for desktop, Web, and SDK, allowing developers to experience, reproduce, and extend it.
Reflex Agent project address: https://github.com/APUS-AI-Lab/APUS-Reflex-Agent
Reflex decision model: https://huggingface.co/apus-ailab/APUS-Reflex
This article was provided by APUS AI and reposted by QbitAI with authorization; the views belong to the original author.