Memory

A long-term memory box for DeepSeek Harness settings: type it once and every new session carries it. 0.3.0 can also expand a rough note into a five-section requirement sheet that keeps every requirement of yours verbatim, marks anything the model had to invent as 推断 and deletable, and lists the dimensions you left open instead of deciding them for you.

Adds a Chat mode beside Work with a user-defined persona, long-term memory extracted after each turn and recalled by topic, and stickers it can save and send.

Iter Memory: it remembers my own state, not the facts. A level that drifts back to baseline by itself, plus an occasional line I choose to say only when that level allows it. When it stays quiet, nothing is spent.

Local knowledge base for DeepSeek Harness: distills workspace and chat-attached PDF, Word, Excel, Markdown, TXT, CSV and HTML documents into Markdown notes, with a VS Code-style viewer, per-workspace access grants and agent tools to search, read and add notes.

Learns a user's coding conventions from instructions and corrections, stores them as confidence-scored rules in editable Markdown, and injects them as reference data into every session. Rules that decay out of range or contradict a newer rule stay visible for review instead of being applied.

Cross-session memory for DeepSeek Harness: a three-tier store (behavior notes, deep details, session archive) plus a 600-character core card injected every step, work/companion/learning state lines, multi-person and multi-group records with visibility partitions, session-scoped suppression or deletion of memory fragments, summaries and fact candidates that can route to a separate model, and backup plus self-check tools. Data is plain text under the home directory, and entries can carry images and audio stored by content hash.

Injects your home folder — identity, house rules, recent diary and ledger — into every new session before anyone speaks, and keeps that home writable so the agent can amend its own persona. Does nothing until the home folder exists (default dsh-home under the user home, or set home in the config).

Warns the session to flush what matters before context compaction kicks in, and rolls a long-term ledger into a frozen archive with a digest of title plus up to three lines per moved section. The roll-up stays off until home is set.

Meta's Muse Code CLI inside DSH: a composer switch starts muse serve, a right-sidebar panel and five chat tools (muse_open, muse_prompt, muse_read, muse_status, muse_interrupt) share one Muse session, approvals are decided only in the panel, and a local spend guard hard-stops at $10 by default. Windows only; needs the Muse Code CLI installed and signed in. The $10 cap is a local estimate for the muse serve this plugin starts.

Forced collaboration: a composer picker sets 0–6 sub-agents, and while it is on every hands-on task must be split across exactly that many child agents through a collab_fanout tool that has no count parameter, then merged by the main agent.

Turn-granular context memory for DeepSeek Harness.

Agent Memory plugin for DeepSeek Harness: local, file-per-memory long-term memory with human approval.

A philosopher's way of thinking made executable: dialectics and practice-testing to govern hallucination and long-session drift, so done is a verifiable fact rather than a claim.

Disk-backed long-term memory for DSH: a SQLite FTS5 vault that captures every conversation turn, injects relevant past turns into new sessions within a per-turn character budget, groups sessions into resumable conversation files, and ships a web panel for search, paging and capacity.

Connects the Drafta notes app to DeepSeek Harness over MCP, so the agent can search, read and write notes in your local Drafta library.

Workspace memory for DeepSeek Harness ported from ZCode: one Markdown memory directory per workspace (a MEMORY.md index plus one file per fact), injected into each main session and refreshed by a background extraction request after completed turns; a sidebar page browses the stored files read-only.

Tide Memory: a long-term memory that forgets on purpose. It keeps what matters and lets the rest fade. Whether an entry enters the context is decided by an evidence score C = s/(s+2): below the threshold (0.35 by default) it never enters. Events decay by half-life (30 days, or 7 days when they carry emotion); facts never decay; superseded entries are kept and marked stale instead of deleted. Recalled lines do enter the context and therefore do consume tokens — small, but real. No embeddings, no network calls, no cloud.

Local knowledge base for DeepSeek Harness: six tools to ingest Markdown, plain text and PDF files, search them with BM25 or hybrid BM25 plus optional local embeddings, and answer with cited source passages. Runs offline with no database server. Requires Node.js 18+ and Python 3.9+; PDF ingest needs pymupdf.

Local knowledge base for DSH: one Markdown file per note, with tags, hierarchical folders, full-text search, attachment opening via the system default app, zero-dependency Office text extraction, and citation into the conversation. Built for host DSH 0.2.0-rc.2; on Windows the default root is E:/dsh-kb.

Local-first knowledge base for DeepSeek Harness: imports Word/PPT/Excel/PDF (text or scanned), EPUB and MOBI/AZW3, builds Markdown parts with page anchors, OCR quality/confidence and column-aware PDF reading order; adds an LLM-maintained wiki layer (schema.md rules, auto index.md, append-only log.md, typed notes with sources and backlinks), adds source-trust labels (high/medium/low) with trust-aware ranking and low-trust warnings, and supports ranked search, cross-KB query, precise reading/editing, and md/jsonl/zip export. MOBI/AZW3 need Calibre; legacy .doc/.ppt need Word/PowerPoint (Windows) or LibreOffice (Linux); scanned-PDF OCR downloads models on first use.

DeepSeek Harness 本地长期记忆(单包单入口):JSONL 存储 + 本地向量检索(BGE-M3)+ 四个模型工具 + 自动召回注入 + 管理面板。

Layered long-term memory for DeepSeek Harness with automatic model distillation, source-linked encrypted session logs, and guarded compaction.

Cross-session memory as plain Markdown per workspace: weighted injection, bi-temporal invalidation, TTL review with stale markers, credential redaction, error and session retrospectives, SOP skills, and approval-gated self-improvement proposals.

Deterministic goal-completion gate for DSH: a contract-first AC grammar with named failable check commands, a gate that re-runs every check itself (zero model trust), verdicts bound to a tree digest so file changes void passing verdicts, and a hard deny on false completion at update_goal via tools/pre-execute with a false-complete counter that blocks after two catches; scope-aware task-level claim handling (strict or goal-only) keeps multi-task Agent Teams loops from deadlocking.