DeepSeek Harness Plugins · Memory
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
Adds a project context graph to DeepSeek Harness, auditing and injecting relevant code dependencies, PDF sections, and structured context before each model request.
A local-first context engine for DeepSeek Harness that tiers memory by recency: preserving current tasks verbatim, reversibly compressing recent history, and folding older history into engineer handoff notes with full recoverability.
Adds cross-workspace session migration to dsh, copying an idle session with full logs, model, and presets to a target workspace while keeping or archiving the original.
A typed, inference-capable ontology memory for DeepSeek Harness where the agent declares domain vocabulary and every assertion is constraint-checked, with automatic derivation via transitive, symmetric, and inverse relations.
Copy a session reference with one click and read the full conversation in any other DSH session, eliminating the manual export, transfer, and log-decoding dance.
A sticky-note plugin for DSH WebUI: jot down text, to-dos, and images with drag, resize, collapse, archive, and session binding.
Auto-detects local tools, paths, network, and disk; persists a cross-session profile; and injects a low-cost summary into every turn's system prompt to cut redundant probing and token usage.
A local Node-based knowledge base for the 12 principles of animation, providing memory and reference tools in sessions with no network dependencies.
Turns the model context into a visible, editable, effectiveness-scored first-class object with token accounting, live refresh, and zero runtime dependencies.
A DSH Web UI plugin that converts AI answers into multiple-choice, true/false, or mixed questions for instant self-testing, answer review, and building a personal question bank with a study list.
A local-first DSH plugin that auto-distills agent conversations into reusable Wiki pages and Skills with progressive discovery and on-demand recall.
Wires the Memtrace code intelligence graph into DeepSeek Harness with structural search, blast-radius analysis, temporal evolution, and 27 agent skills, all running locally.