Memory & Contextdsh-plugin

dsh-layered-memory

dsh-layered-memory is a DeepSeek Harness plugin for cross-session long-term memory. Its README describes background L0 capture, L1-L3 distillation, automatic recall injection, separate personal and work memories, SQLite storage, optional embeddings, and zero-config operation.

Stars
13
Forks
3
Open issues
2
Last push
Sep 6, 2026
Latest release
v0.10.0
24h Growth
+0
7d Growth
+0

Installation

dsh plugin --profile web add dsh-layered-memory

Overview

dsh-layered-memory is a DeepSeek Harness plugin for cross-session long-term memory. Its README describes background L0 capture, L1-L3 distillation, automatic recall injection, separate personal and work memories, SQLite storage, optional embeddings, and zero-config operation.

  • Background layered distillation pipeline: L0 conversation capture, L1 atomic memories, L2 scene integration, and L3 persona distillation.
  • Automatic recall injects relevant memories into context before each model step, with length and time budgets for injection.
  • Separates personal and work memories, with session-level memory modes such as daily, work, smart, and off.
  • Memory tools include memory_search, conversation_search, and memory_read_scene.
  • Uses SQLite storage and JSONL fact sources; vector search is optional and disabled by default (pure FTS).
  • Embedding sources include off, remote OpenAI-compatible /embeddings, and local ONNX CPU models.
  • Settings include a memory workbench with overview, memory library, automation, insights, and maintenance sections.
  • A cost dashboard records distillation LLM token costs by provider/model in SQLite.

Related plugins