dsh-memos-bridge

DeepSeek Harness agents are stateless by design: once a conversation ends, its context vanishes. dsh-memos-bridge closes that gap by bridging the MemOS persistent-memory service into the agent over MCP, purely as a configuration layer—no Harness source code is touched. You mount the standard MCP client, point it at a stdio server, and the agent instantly gains cross-session memory read and write capabilities.

With the bundle active, the agent can invoke 16 mcp__memos__* tools spanning the full memory lifecycle: creating memories from text, documents, or conversation messages; performing semantic search across memory cubes; inspecting, correcting, or deleting individual memories; creating and sharing cubes; engaging in memory-enhanced chat; and starting or stopping the asynchronous memory scheduler. The runner performs a hard preflight check against Neo4j at boot, warns clearly when the LLM or embedding gateway is unreachable, and reconnects automatically once the stack recovers—no manual GUI restart required.

This bundle is best suited for developers who already run the MemOS docker stack (Neo4j, Qdrant, and the memory API) along with a reachable LLM/embedding gateway, and want their agent to possess truly persistent, searchable, and manageable long-term memory.