DeepSeek Harness Plugins · Memory
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
Unified memory manager for AI Agents with sliding-window short-term recall, hybrid vector-search long-term retrieval, layered distillation, and lifecycle control - local-first, zero cloud dependency.
Session recycle bin for DeepSeek Harness with sidebar quick-move, batch restore or permanent delete, and live-session protection.
A local evidence ledger for DSH agents that appends facts, tests, decisions, and failures to a workspace JSONL file with filtering, full-text search, and no network calls.
Adds persistent long-term memory to DeepSeek Harness via a self-hosted Mem0 server, automatically recalling relevant facts before answering and distilling each completed turn into stored memories.
Indexes large files and web pages into a local knowledge base, blocks context-overflow data floods at the gate, and runs analysis in a sandbox returning only results to minimize context usage.
Import existing Codex or Claude Code conversations into DSH, auto-bind the workspace, and continue the dialogue with the DSH Agent.
Hybrid memory plugin combining Hermes frozen-snapshot L1 with Noema searchable knowledge base L2, featuring prefix-cache optimization, Chinese FTS5 search, cross-tool import, and local-only storage.
Mines local DSH session logs and git history into Markdown and self-contained HTML usage reports with zero network, zero config, and optional path redaction.
Ports familiar slash commands from Claude Code and similar tools to DeepSeek Harness with 17 verified commands and mandatory adversarial trust review before any plugin recommendation.
Turns local papers and spreadsheet files into a Zotero-style floating drawer in DSH Web for reading, note-taking, in-place editing, and chart visualization.
Attach one custom system prompt per workspace in DeepSeek Harness, auto-captured into each new session, manageable via a header dialog or Agent tools.
A DSH context-enhancement plugin for long sessions that uses tool summarization, zone-based compression, and an independent checkpoint sidecar to reduce information loss after /compact.