dsh-memory-plugin

Every new session in DeepSeek Harness starts from scratch: your preferred agent, go-to model, active project, and recent tasks all need to be re-stated. dsh-memory-plugin exists to fix that. It quietly accumulates a picture of how you work across sessions, so the next time you open a terminal, your assistant already knows your preferences.

The plugin tracks several dimensions automatically: call frequency of common tools such as read, write, edit, glob, and grep; your preferred agents, LLMs, language settings, and code style; active projects with their tags and access history; and session-level details like discussion topics and completed tasks. A built-in recommendation engine uses this accumulated data to suggest the best-fit model and agent configurations on demand. All memory is persisted as a local JSON file with import, export, and clear-Memory API support, giving you full ownership of your data. A web visualization viewer lets you browse your accumulated memory through charts, timelines, and summary cards, and eight real-time recommendation metrics are available for inspection.

The plugin is well suited for developers who use DSH frequently and juggle multiple projects or toolchains. If you want your AI assistant to gradually adapt to your workflow while keeping every byte of data local, cloud-free, and deletable on demand, this lightweight and transparent memory layer is a practical fit.