Introduction

DeepSeek Harness starts with a fresh session by default. This means you need to repeatedly explain your identity, work preferences, communication style, and confirmed decisions. The dsh-memory-migration plugin provides long-term memory for native DSH: a key-value store for “who you are and how you work,” recalled in later conversations based on question matching.

Plugin Positioning

This is a DeepSeek Harness plugin maintained by tieveto666-code, intended to address the lack of long-term memory in native DSH. It does not modify DSH source code, and implements memory persistence and retrieval through local storage and a BM25 scoring mechanism.

Core Features

  1. Key-value storage: Maintains local records of identity, preferences, and working methods.
  2. Intelligent recall: Based on the BM25 algorithm, retrieves entries matching the current question in later conversations. Matching logic includes containing the full key name or content-based semantic matching (when a specific threshold is met).
  3. Memory migration: Supports extracting core memory pairs from official exports (ZIP or JSON) of ChatGPT, Claude, Gemini, or DeepSeek.
  4. Settings interface: Provides a visual key-value table, bulk operations, search, and a global toggle.
  5. Non-invasiveness: Runs as an out-of-tree Cordis plugin without forking or patching DSH source code.

Installation and Enablement

Prerequisites: It must run under the Web profile, and Node.js version must be 22.19+ or 24+.

Installation command:

dsh plugin --profile web add github:tieveto666-code/dsh-memory-migration

After installation, start dsh --profile web and find Long-term memory in settings.

Typical Usage

  • Maintain consistency: Keep identity, communication style, and hard preferences unchanged across sessions.
  • Reuse methodologies: Reuse established methods and constraints for specific task categories (such as code review or SQL writing).
  • Avoid repeated discussions: Confirmed decisions and their rationale are recorded, so new conversations no longer need to relitigate them.

Notes and Limitations

  • Data storage: Data is stored in the local $DSH_HOME/memory-migration/ directory.
  • Migration limits: The migration feature extracts only core memory pairs (up to 10), not complete conversation records. Migration filters out one-time task conclusions, schedules, secrets, and the assistant’s personal opinions.
  • Privacy: Parsing, unzipping, extraction, and storage are all performed on the DSH host. Data is not sent to third-party services, and no telemetry is collected.
  • Dependencies: If you use the migration feature, ensure DSH is configured with a DeepSeek model (used to extract core pairs). If not configured, only existing enabled facts in the file are imported.

Uninstall

When uninstalling, remove only the plugin layer. To delete the local data directory, run the command manually:

dsh plugin --profile web remove @tieveto666-code/dsh-memory-migration
rm -rf $DSH_HOME/memory-migration/

Summary

By introducing locally persistent memory, this plugin addresses the fact that DSH resets each session, making it suitable for scenarios that require long-term consistency in persona or workflow. For more details, see the plugin directory or the source repository.