DeepSeek Harness Plugins · Memory
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
A DSH plugin for self-improving AI agents that enables continual learning through persistent memory, periodic review, cross-session knowledge sharing, and automatic rollback on failure.
A plugin porting TencentDB Agent Memory into DeepSeek Harness, featuring a four-layer memory system with conversation capture, structured memory extraction, and automatic recall injection for dynamic context.
dsh-obsidian-assistant is a DeepSeek Harness plugin for operating local Obsidian vaults with dual-channel access, enabling search, note management, and batch operations.
dsh-prompt-manager transforms DeepSeek Harness system prompts from a black-box string into a visible, configurable, and auditable pipeline for fine-grained management and learning of the prompt assembly process submitted to the server.
The DSH Archive Manager plugin provides Codex-style conversation archive management for the DeepSeek Harness web interface, supporting viewing, searching, previewing, restoring, and deleting archived chats.
Task-isolated long-term memory for DeepSeek Harness, storing memories in per-task vaults to ensure data isolation.
GitLearnOS is a native DeepSeek Harness plugin that provides a learner-owned Git memory system to record learning events and guide independent learning.
Automatically generates a sanitized ZIP evidence package with pre-failure state, project info, and environment when a tool or agent fails, for local diagnostics.
dshx is a CLI tool for managing MCP servers, skills, and memory in DeepSeek Harness, featuring one-command add/remove, connection tests, secure configuration, and migration from other environments.
A DSH web plugin for per-host personalization with custom instructions, local long-term memory, and reply-tone presets applied to every conversation on the machine.
dsh-session-link is a memory plugin that injects session context snapshots across dialogues via deep links, enabling cross-conversation referencing.
Unified fleet-wide agent memory system with Obsidian Vault storage and SQLite FTS5 indexing, providing full lifecycle management of knowledge including ingestion, promotion, adjudication, and decay/retention for cross-agent sharing.