DeepSeek Harness Plugins · Memory
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
Provides an Obsidian-based knowledge OS for coding agents with daily sealing, distillation, promotion, and deterministic short-noun page lookup.
A learning-mode plugin that turns DSH session concepts into structured learning cards, building a continuous global knowledge loop locally without affecting the primary agent.
A local-first Token observability, budgeting, and trajectory audit plugin for DeepSeek Harness, offering multi-dimensional usage tracking, public rate estimation, and on-demand AI-driven session analysis.
Bookmark DSH assistant replies with notes and tags, manage them in a cross-session center panel, and export to Markdown in one click.
Cross-session memory: the model saves facts, decisions, and preferences as Markdown while working, auto-injected into new sessions with on-demand topics and optional background distillation.
A local-first archived chat center for DeepSeek Harness, offering full-text search, read-only previews, version history, and a secure recycle bin while keeping all data strictly local.
dsh-zotero is a read-only Zotero extension for Agent research workflows that enables direct literature search, evidence extraction, and citation generation from your local library.
Seamlessly connect DeepSeek Harness to your local Obsidian vault, enabling agents to search, read, write, and manage Markdown notes directly via the file system without extra servers.
A layered distillation memory plugin for DeepSeek Harness that automatically captures, distills, and injects relevant context before model inference.
Quickly view, search, and jump to all your sent messages in ultra-long conversations, featuring full history retrieval and one-click auto-scrolling.
Automatically distills your explicit corrections, constraints, and workflows in DSH into reviewable native Skill drafts, saving them for future reuse only after your confirmation so your experience never fades with the session.
Automatically logs all-mode tool failures, normalized and deduplicated into skill logs, to help models reduce future failures.