dsh-memory-migration

Every DSH conversation starts from a blank slate. Your preferences, standing methods, and confirmed decisions either get re-explained every round or sit unused inside another product's export file. dsh-memory-migration adds a compact long-term memory layer to stock DSH: a local key/value store that captures who you are, how you work, and which constraints are non-negotiable, then recalls only the entries that match the current question rather than dumping the entire history.

On the everyday side, the Settings panel lets you add, inline-edit, search, paginate, and bulk-delete memory rows, with a global switch to enable or disable recall. The matching pipeline first checks for exact key containment, then falls back to BM25 scoring over CJK unigrams and bigrams plus Latin words; injection is capped at eight keys and sixteen entries, and off-topic questions receive nothing. The optional migration path imports a ChatGPT, Claude, Gemini, or DeepSeek official export; using the DeepSeek model already configured in your DSH profile it scans recent substantial conversations and distills at most ten core pairs covering identity and hard preferences, methods for a task class, recurring constraints, and confirmed conclusions with rationale, while merging near-duplicates and skipping pairs similar to memories you already have.

The plugin does not fork or patch DSH source, does not turn old chats into sidebar sessions, stores everything under the local memory-migration directory, and sends no telemetry or export data to any third-party service. It is aimed at users who want to keep communication style and hard preferences across new DSH chats, who want to reuse standing methods and constraints for a class of work, or who are migrating from another AI product into DSH.