Plugin

Long-term memory injected at session start: preferences/facts/summaries/knowledge in global and per-project scopes, budgeted recall via `agent/pre-step`, durable JSON store.

Enhanced auto-compaction defaults for DeepSeek Harness agent presets: apply tuned threshold/retention budgets and per-model policies with one command, idempotent with backups and rollback.

Brings your existing Claude / ChatGPT history into DSH: instant zero-token import of the memories.json inside a Claude export, local Claude Code transcripts with no export needed, and conversations distilled through the model you already configured in DSH. Every memory carries verbatim evidence checked against the source by code — a paraphrase is dropped and counted, and evidence found only in the assistant turns is always routed to review rather than stored.

Conversation history recall for DSH: three-layer (literal/fuzzy/semantic) retrieval over every past session's original text, fully local & offline — AI never forgets what you told it. One-command install via `dsh.bundle.patch`, semantic layer runs in a worker thread.

Agent tools for a self-hosted mem0 server: add, search, get, update, delete, history, reset and status memories, with a settings-panel configuration card.

Cross-session plaintext memory: deterministic BM25 keyword recall (no embeddings), memory_save/list/search/confirm/forget tools behind a human-confirm gate, and global + project-scoped JSON stores that follow your git repo.

Provides local and remote knowledge bases with project- and session-scoped recall, controlled write-back, and an embedded Web management console.

Layered memory pipeline for DSH: auto-distills conversations into atomic facts, scene summaries and a persona profile (L0–L3), with hybrid BM25 + vector retrieval, chat/work family separation, and context injection before each model step.

Automatic, local-first cross-session memory for DeepSeek Harness: remembers user preferences, project decisions, conversations, and tool results; verifies recalled information against original messages before answering.

Minimal-compatible first turn for Web root sessions: starts with persistent bash and str_replace_editor, removes automatic workspace and skill context, then restores the selected preset after the first tool call or reply; includes a persistent composer toggle.

Deterministic context compaction: prunes conversation atoms in reference-graph topological order with no LLM calls during compaction, and keeps pruned atoms recallable by marker.

Imports chat.deepseek.com web conversations into DeepSeek Harness as resumable sessions, from a settings-page UI.

Project and session organization with multi-folder groups, attention summaries, grouped chat search, native folder picking, pinning, favorites, ordering, and an archive center.

DSH WebUI session query navigation with a faint conversation-edge rail, a hover list of real user queries, and click-to-jump back to the matching bubble.

Message review and quote chips for conversations: multi-model cross review (code/reply) with severity grading and summaries, review history, and host-persisted preferences.

Injects one fixed system-prompt section so the agent thinks and replies in Chinese, regardless of the user's language.

Auto-repairs the messages array after a tool-dispatch crash (orphaned tool_calls / tool messages), preventing 400 INVALID_REQUEST session lock-ups.

Full-text global search for the DSH Web GUI: a sidebar search box that searches history sessions (via the sessionQuery index, with a per-session scan fallback) and settings panels, highlights matches, and jumps to the matching session or settings card.

Delete conversations with a confirmation dialog and manage archived sessions (archive/unarchive) in the DeepSeek Harness web UI.

Bidirectionally synchronize sessions, workspaces, settings and plugins between the DSH web instance and the DSH Desktop instance.

Sub-agent matrix swarm that routes heterogeneous tasks to the most suitable model from an OpenRouter-like gateway plus cfgpu.com/llm/square, dispatches each via in-process subagents or direct LLM calls, and tracks per-model token consumption with a feedback-driven ranking.

Ollama Cloud native chat adapter: direct NDJSON translation of the Ollama /api/chat wire format, model discovery with context windows and capabilities, and web search/fetch providers under the ollama-cloud id, plus a Web settings card.

Writes live provider model lists into DSH settings and shows 5h/7d plan windows or metered balance for the current session model.

macOS Keychain credential provider that replaces the local-file provider and uses a signed, notarized universal helper.