Plugins

人格 + 情感 + 记忆的 DeepSeek Harness 插件：把「一个会记得你的角色」接进 DSH。

Cross-workspace long-term memory for DeepSeek Harness: one machine-local store shared by every workspace, a compact index injected at session start, and a searchable tool for the details.

Cross-session memory for DeepSeek Harness. Local memory store by default, optional GitHub-backed sync (git), with primary/child roles for multi-device setups.

Cross-model continuity for DeepSeek Harness: memory, skills and rules live in local files, so judgement survives session and model changes. Not about storing more memory — about using it well after the model changes.

Shows how many GPUs on the dsh-remote host are free in the conversation header, and per-GPU utilization, memory, temperature and the owning user in a right-sidebar panel. Adds a remote_gpu tool that returns the same snapshot.

Connect DSH to a local BiliMemory backend for Bilibili search, collection, status and login tasks, A-Memorix shared memory health checks, plus QQ link status, QQ user allowlists and QQ watcher restart. The repository includes a Windows one-click installer that installs or repairs MaiBot A-Memorix when missing, then configures local-bge-small-zh, embedding and shared memory. It also provides install-memory.cmd for memory-only repair. Requires Windows, DSH Desktop 0.2.1-rc.2+, Node.js 22.19+, Python 3.12, Playwright/Edge, FFmpeg, a running BiliMemory API at 127.0.0.1:8765, a local embedding service at 127.0.0.1:11435, a Bilibili login, and MaiBot/NapCat for QQ. Voice, RVC and cover features are excluded.

Agent memory with a forgetting curve: entries gain weight when mentioned, shrink when forgotten, and pinned entries never decay. Six tools and a settings panel.

Long-term notes and daily journals kept per workspace in a hot/cold archive, each tier with its own index and rebalanced once a day; read and written through seven memory_* tools, with a settings page for per-workspace options.

Long-term memory kept as blocks: one event per block, written to a local append-only event log with a block view and derived indexes, so a full history stays recoverable while only the matching blocks reach the context. The model saves entries with memory_remember and pulls the relevant few with memory_recall, plus one automatic recall pass before each step. Data stays in ~/.beme; no external service and no API key.

Turns archived sessions into durable memory instead of dead disk. Distils each archive into reusable skills, a fact ledger with ADD/UPDATE/NOOP decisions plus validUntil and supersededBy so stale facts retire instead of piling up, and topic clusters by single-linkage similarity. Also digests sessions that are still in progress through a read-only seq watermark, leaving the session log byte-for-byte untouched, and can self-check the growing memory to merge duplicate facts, retire stale ones and consolidate duplicate skills — measuring only unless you opt in, and never hard-deleting, only superseding.

Local long-term memory for DeepSeek Harness with SQLite storage, automatic extraction, semantic recall, corrections, reliable forgetting, and an in-app summary dialog.

Cross-session long-term memory for DeepSeek Harness: pinned rules injected every turn, the rest recalled by IDF-weighted keyword evidence, fact-key dedup, plain JSON storage, and a settings page to edit or delete entries.

DSH-elysia — shared Markdown memory for every DSH session: persona, requirements, corrections, lessons, methods, requests, sessions, a method library and ingested documents, mirrored to a second disk with version snapshots and a cross-process write lock; entries move between the main file and a searchable archive by hit frequency and repeated emphasis, and retrieval is keyword co-occurrence plus BM25. Ships a skills library (SKILL.md playbooks with optional guarded run.mjs execution), a sticker library that sends a fitting sticker at the right moment, and — new in 1.10 — Memory Packs: a bundled, opt-in character memory (packs/elysia: 61,903-char identity + 2,482-char persona distilled from 1,696 official voice lines and 112,175 dialogue lines). Packs are off by default, back up before writing, only touch identity/persona, and can be uninstalled to restore what was there.

Six-tier memory plugin that captures each conversation turn, injects a stable memory block as a user snapshot, and compacts the host's own history into a single placeholder without calling a model.

Sidebar panel management for DeepSeek Harness: a grouped Panels launcher collects global panel rows and third-party plugin entry rows into one searchable menu — the full panel inventory (pinned and unpinned, unpinned open via temporary-pin peek), collapsible group memory, in-menu search with hit highlighting and keyboard navigation, Alt+1..9 direct activation of the Nth visible item, and hover pin/top quick-actions. A Ctrl+K command palette mirrors the inventory with quick commands, a Recent group of last-used panels (MRU, persisted), and named layout presets — save your whole arrangement (pins, top pins, folds, whitelist, icon-only, compact footer) and re-apply it in one click. v1.0.0 adds custom panel groups: define your own groups (inline rename, member picker in Settings) and they render as first-class groups with accurate counts in the menu and palette. Plus a keyboard-reachable right-click context menu, icon-only panel rows, top-pinning with drag-to-reorder, a compact footer with granular sub-toggles, and a Settings page with height sliders, a locale-safe whitelist, pin management, and settings export/import.

Long-term memory backed by a local SQLite store: project-scoped memory auto-injection with junk filtering, rule-based or model-assisted memory extraction from recent turns, write-side dedup and feedback-weighted ranking. The Python bridge runs on $NEXUS_PYTHON or the DSH-bundled interpreter at <DSH home>/dsh-runtimes/dsh-primary-runtime/dependencies/python/ (Python 3.8+).

Layered cross-session memory kept as plain YAML: facts carry a version history instead of being overwritten, a briefing is injected at session start, and a nightly pass consolidates and shards by topic past a threshold. Zero runtime dependencies, so it also runs inside the dsh Android app (verified there).

Configurable personas, separate reply generation, and evidence-backed memory with review and correction for DeepSeek Harness.

Choose API hostnames from configured providers or add them manually to route matching DSH backend Fetch/Undici requests through an existing local HTTP(S) proxy, leaving other routes unchanged; optional in-memory records show actual dispatch paths, first-content wait, token throughput and usage without storing prompts, responses or API keys.

Fills in a Chinese description for every installed DSH plugin that ships only an English one, at startup and again after each plugin change, and never rewrites a plugin name. Plugins that already export a locale folder show the new text on refresh; the rest are patched too — the in-memory layer shows Chinese immediately, and the on-disk locale pack it writes needs one restart.

memcue long-term memory for DeepSeek Harness (dsh): the model searches and writes its facts itself

深鲸记忆：跨会话热记忆（每轮注入 MEMORY/USER + memory 工具读写）+ 记忆体检、瘦身归档、记忆包、诊断报告、文档只读。本地明文、零依赖、不上传数据。