ByteOpenViking Core Contributor Open Sources LoopX: Allows Agent to Run Continuously for 200 Hours Without Forgetting or Deviating

According to DeepBeat Monitoring, Huang Ruiteng, an AML Senior Machine Learning Engineer at ByteDance and a core contributor to OpenViking, has open-sourced LoopX. It is a control system for long-range agents that allows agents like Codex and Claude Code to continue their original objectives after tasks spanning multiple days and interruptions.

LoopX has publicly released two real mission trajectories spanning 220.7 and 272.9 hours. During this time, which included multiple rounds of execution, waiting, human judgment, model switching, and task recovery, the agents were able to reacquire their current objectives, existing evidence, and next steps.

LoopX abstracts objectives, to-dos, permissions, evidence, and waiting conditions out of the model context. Agents take small steps at a time, validating the results before writing back to the state. Whether switching sessions, models, or restarting the program, they can continue from the latest progress.

Developed by Huang Ruiteng, OpenViking is responsible for storing and retrieving agents' memories, data, and skills; LoopX manages where tasks have been done, what the next step is, and when human judgment is needed. The former is akin to long-term memory, while the latter is more like a project manager and an executable board. Currently, LoopX has been used for automatic code issue fixing, AutoML experiments, and long-term research.