dsh-model-memory

When diving deep into DeepSeek Harness (DSH), precisely controlling model reasoning processes can often be a headache. Manually configuring thinking levels is not only tedious but also hard to maintain consistency across multiple sessions. The dsh-model-memory plugin is designed to solve this pain point, encapsulating complex configuration logic directly into the native interface, allowing you to fine-tune model behavior without cumbersome operations. The plugin boasts three core capabilities: First, it supports one-click configuration of model thinking levels (Low/Max, etc.) and features cross-session memory, automatically recalling your preferences for different channels so that new sessions are seamlessly populated. Second, it introduces flexible channel-level retry strategies, supporting default retries, infinite retries, and custom counts to ensure request high availability. Finally, it pursues a pristine experience by removing redundant interfaces and utilizes a symlink mechanism to achieve upgrade immunity, ensuring your configurations remain safe even after global environment updates. For AI researchers, developers, and advanced users who demand ultimate control, this plugin is an essential tool for streamlining workflows. It allows you to enjoy the native fluidity of DSH while acquiring deep customization flexibility, making it the ideal choice for building a personalized AI workspace.