AI Agent Hub
Back to plugins
🧠

dsh-workspace-memory

Memory Updated 2026.09.03

Run the following command in DeepSeek Harness:

dsh plugin install lgquan/dsh-workspace-memory

Paste the following prompt into your AI chat to install this plugin:

Run dsh plugin install lgquan/dsh-workspace-memory in your terminal to install; the source repository is available at https://github.com/lgquan/dsh-workspace-memory

About this plugin

Sessions in DeepSeek Harness are isolated by default—the technical decisions, module contracts, and user preferences confirmed in one session vanish the moment you open the next; with multiple projects running in parallel, memories bleed into each other. dsh-workspace-memory isolates long-term memory by workspace, giving every project its own memory scope while sharing global preferences across projects. It eliminates context loss and cross-project contamination without requiring a vector database or an Embedding service.

After each conversation the plugin buffers, distills, and organizes memory on schedule, keeping only durable artifacts such as architecture decisions, module responsibilities, verified conclusions, and stated user preferences. Ephemeral chat, one-off task details, and credential-shaped content are rejected from persistence and re-sanitized before any memory output. On the retrieval side, every Agent step receives a size-capped stable summary; the first step retrieves relevant structured memory for the current question, and later steps can call memory_search for a second pass. On the correction side, entries can be deleted by ID, explicit user corrections supersede old facts, contradictions with insufficient evidence are preserved as visible conflicts rather than silently overwritten, and pre-revision content remains available for audit. Memory status, revision history, conflict records, and a recycle area for deleted projects are all inspectable in the Settings panel.

It is built for developers who want multiple sessions of the same project to share context, who want the standard Agent and the voice Agent to draw on the same project memory, and who want long-term memory to be inspectable, correctable, and deletable rather than a black box. Once DeepSeek Harness and the web profile are in place, a restart after installation surfaces the memory panel in Settings; when paired with dsh-voco, voice main sessions and delegated sub-sessions automatically share the same project memory scope as long as they work in the same directory.

Use Cases

  • Multiple DSH sessions of the same project share confirmed tech decisions and module contracts
  • Standard, background, and voice Agents draw on the same project memory scope
  • Inspect, correct, and delete long-term memories while preserving auditable revision history

Best For

  • Developers who switch between multiple DSH sessions within the same project
  • Teams that want standard and voice Agents to share consistent project context
  • Users who require long-term memory to be inspectable, correctable, and deletable rather than a black box