dsh-echo-memory

dsh-echo-memory tackles a very concrete pain point: every new session, the agent has no memory of your preferences, project constraints, or decisions you already agreed on, so you end up re-explaining yourself. This plugin lets the agent naturally capture that information during conversation and automatically recall it in later sessions. Only when relevant, one or two items at a time, no context flooding.

The design is deliberately minimal. All memories live in a single local memory.json file that is human-readable and atomically written. Retrieval uses BM25F scoring with title and tag weighting; ten thousand entries are searched in single-digit milliseconds. No vector database, no external API key, no local database required. On the interaction side, a draggable floating dot in the bottom-right corner lets you drag selected web text directly into memory; AI-suggested items pop up as confirmation cards; and a tombstone recycling bin guards against accidental deletions.

This plugin fits DSH users who want their agent to behave like a partner with persistent memory, but do not want to hand that memory to any external service. You prefer a purely local, auditable data model where you can open the JSON file and read it at any time. You switch between multiple projects and want project-scoped memories to stay isolated while global preferences remain shared across them.