dsh-memory: A DSH plugin for Long-term Memory in DeepSeek Harness
qwert702/dsh-memory
Long-term memory for the DeepSeek Harness Web GUI: project+global stores, auto extraction/injection, small-model consolidation, and an Obsidian-style link graph.
dsh-memory is a long-term memory plugin for DeepSeek Harness, bringing project-scoped and global memory stores to the Web GUI. It automatically extracts conversation highlights, injects relevant memories on demand, and periodically consolidates them with a small model, so the model retains key information across sessions. It includes an Obsidian-style link graph with search, editing, and visual browsing. With it, your DSH conversations no longer lose context, and your AI assistant remembers things from long ago.
How to Install
dsh plugin --profile web add dsh-memory- Category
- Memory & Context
- Platform
- DSH Plugin
- Author
- qwert702
- Distribution
- Plugin
dsh-memory Key Features
dsh-memory Repo Summary
What Does It Do?
dsh-memory is a DSH plugin for DeepSeek Harness that provides long-term memory for the Web GUI, with project-scoped and global memory stores. It solves the problem of lost context across conversations, letting the model retain key information between sessions. Core capabilities include automatic extraction of conversation highlights, on-demand injection of relevant memories, periodic consolidation using a small model, and an Obsidian-style link graph visualization.
Core Features
- Two-scope storage: per-workspace project memories (keyed by a hash of the session cwd) plus a global store, saved as plain JSON under
~/.dsh/memory/with atomic writes. - Auto injection: a global system-prompt assembly listener appends top-K memories (ranked by freshness, reinforcement, and degree) as dynamic context; failures never affect the assembly.
- Auto extraction: on turn end, new events are replayed as a transcript and distilled by the model into JSON candidates; near-duplicates (bigram Jaccard ≥ 0.65) reinforce instead of duplicating.
- Periodic consolidation: every N extracted turns (default 20) or on manual trigger, the model proposes merge/link/archive/retag operations that are id-validated and applied transactionally.
- Memory graph: a force-directed canvas with zoom/pan/drag, neighbor highlighting, and a detail card, plus inline edit/archive/link/delete and search.
How to Use This Plugin?
After enabling it in DSH, the plugin automatically listens to conversation events, so no manual intervention is needed. You can adjust settings such as injection toggle, extraction model, consolidation frequency, topK, and max injection characters in the settings file. The memory view tab displays all memory cards with search and graph browsing, and you can also manually trigger consolidation or edit individual memories.
This page is an independent rewrite of the plugin's official README — for authoritative documentation and the latest changes, refer to the source: qwert702/dsh-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.