dsh-meow-memory: A DSH plugin Providing Cross-Session Long-Term Memory for DeepSeek Harness
phant0meow/dsh-meow-memory
Cross-session memory plugin for DeepSeek Harness: seven-layer SQLite store, BM25 retrieval, and per-window dream consolidation to help AI remember users and projects.
dsh-meow-memory is a cross-session memory plugin built for DeepSeek Harness. It consolidates each workspace's memory into a seven-layer SQLite database covering soul, user, project, fact, lesson, topic, and rules. With it, the AI no longer forgets context across windows and sessions—it injects long-term memory on the first turn, performs keyword-based retrieval on subsequent turns, and automatically consolidates memories when a window is idle. It offers tools like memory_remember, memory_search, and memory_project, plus /dream for manual consolidation and per-window skip, making long-term memory management both automatic and controllable.
How to Install
dsh plugin --profile web add meow-memory- Category
- Memory & Context
- Platform
- DSH Plugin
- Author
- phant0meow
- Distribution
- Plugin
dsh-meow-memory Key Features
dsh-meow-memory Repo Summary
What Does It Do?
meow-memory is a DSH plugin for DeepSeek Harness that provides cross-session long-term memory with a seven-layer SQLite store per workspace. It solves the problem of AI forgetting context across sessions and windows, letting you maintain persistent memory, keyword retrieval, and automatic consolidation directly inside the harness without manual memory file management.
Core Features
- Seven-layer memory store: soul/user/project/fact/lesson/topic/rules each in its own SQLite table, with time-prefixed UUIDs so id order equals creation order.
- First-turn long-term memory injection: injects a fixed-format block (full soul, full user, design principles, memory guide) before the first user message; keyword matching starts from the second turn.
- Per-message keyword matching: retrieves fact/lesson/rules/topic based on entry keywords (LLM-extracted or automatic bigram), scored by intersection × idf × coverage × Ebbinghaus decay × importance weight × title bonus.
- Cache-friendly design: a static guide section is registered once in the system prompt, keeping text constant to preserve LLM provider KV/context cache.
- Toolset: memory_remember, memory_search, memory_project, memory_find_similar, memory_read, memory_update, memory_dream, with project anchoring, status management, deduplication, and conflict detection.
- Per-window dream consolidation: triggers when a window is idle for ≥ idleMinutes (default 180 minutes), processing atomic memories, topic memories, and project summaries in rounds, with peak-hour suppression and manual /dream command.
How to Use This Plugin?
After enabling it in DSH, the plugin automatically creates a memory database for each workspace and injects long-term memory and keyword matches into conversations. You can call memory_remember to write memories, memory_search to retrieve, memory_project to view project overviews, or type /dream in the input box to trigger manual consolidation. To adjust behavior, modify parameters like idleMinutes and suppressWindows in the configuration.
This page is an independent rewrite of the plugin's official README — for authoritative documentation and the latest changes, refer to the source: phant0meow/dsh-meow-memory. The plugin is third-party code that runs on your machine once installed; inclusion does not imply endorsement — please review the source before installing.