A
licenseNot graded
qualityB
maintenanceProvides a persistent, four-layer long-term memory backend for DeepSeek Harness agents via 21 MCP tools, automatically weaving relevant context into each turn and ingesting dialogue into scoped stores (facts, evidence timelines, logs, episode summaries). Supports hybrid semantic+BM25+recency retrieval with local ONNX embeddings, AES-256-GCM encrypted field storage, and encrypted backup/restore — all running locally at zero API cost.
56 npm
Apache 2.0