csyangwen/dsh-memory-evolve

dsh‑memory‑evolve – What it is

dsh‑memory‑evolve is a plugin for the DSH chat‑assistant platform. It adds a suite of productivity‑oriented features that let the AI assistant keep long‑term, cross‑session memory, manage to‑dos and skills, coordinate multiple AI “agents”, sync that knowledge across devices, and expose a visual “infinite canvas” for files and notes. In short, it turns a single‑session chatbot into a persistent, team‑like co‑worker.

---

Core capabilities (as described in the README)

Feature	What it does	Typical use case
Long‑term memory (5‑track)	Stores user profile, global facts, project‑specific facts, project logs and daily logs. The AI can read/write these memories on demand and they are automatically injected into future conversations.	You stop re‑explaining project decisions across days or new chat windows.
Memory sync	Pushes the memory branches to a remote Git repository (or a dedicated “shared memory” repo) and pulls them on other machines.	Work on the same project from multiple computers or share memory with teammates.
Todo & project logs	Four separate todo tracks (life, work, project, daily) that are scoped per directory, with due dates and priority. Project logs are automatically created per working directory and tagged with the current Git branch.	Keep track of tasks and progress for several concurrent projects without mixing them up.
Local file search	Simple filename search and content search that returns matching files or snippets to the AI.	Quickly locate prior notes, code snippets, or design assets without leaving the chat.
AI team orchestration	From a “main” session you can spawn additional DSH sessions (e.g., front‑end, back‑end, design) and place them in a shared “room”. The main session can query their status, assign work, and receive results.	Turn a single developer into a small, coordinated AI team.
External AI agents	Dispatch heavy‑weight tasks (e.g., image redesign, code review) to external providers such as Kimi, Grok, Codex, Hermes. The task runs asynchronously, can receive project memory, and its summary is written back to the logs.	Offload long‑running or specialised work while keeping the main chat responsive.
Prompt‑library injection	Store reusable prompt templates (code‑review checklist, testing strategy, etc.) and have the AI automatically inject the appropriate one, or let the user inject it manually.	Enforce consistent workflows across all AI sessions.
Bookmarks & branching	Star any chat turn, create a bookmark with a name/summary, and branch the conversation from that point using DSH’s native branching mechanism.	Manage very long conversations without losing important intermediate decisions.
Infinite canvas	A visual board where you can drop file references, notes, images, audio, video, etc. Items are identified by an ID that the AI can read or write via the de_canvas tool, without loading the full content into the chat context.	Keep all project assets in one place and let the AI fetch them on demand.
Session‑level advisor (reviewer)	An optional hidden reviewer session watches each turn, can raise info, nit, concern or blocker messages, and injects them back as if the user had typed them. Constraints can be set at global, project, session, or reviewer‑only level.	Enforce coding standards, safety checks, or any custom discipline without manual supervision.
Mobile UI & notifications	The DSH web UI works on phones; sessions can be filtered, and the AI can push notifications to Feishu, QQ, WeChat, or an in‑page “bell” UI.	Stay informed while away from the computer.

---

How to get started (quick install)

1. Add the plugin to a DSH profile (e.g., the web profile) with:

   dsh plugin --profile web add github:csyangwen/dsh-memory-evolve
   

2. Restart DSH web – the plugin registers itself automatically via the bundled cordis.patch.yml.
3. Enable the pieces you need in the “Memory Evolve Settings” panel (most features are off by default to keep the AI’s tool list small).
4. Configure optional modules such as memory sync, session advisor, or canvas via the same settings UI.

---

What you get out of it

• Persistent context – the AI remembers who you are, project decisions, and daily progress across sessions.
• Cross‑device continuity – push/pull memory to a Git repo so any machine picks up where you left off.
• Task orchestration – turn a single chat into a mini‑team, with internal sub‑sessions and external AI agents handling heavy work.
• Structured knowledge – todos, logs, skills, and a visual canvas keep all artefacts organized and searchable.
• Safety & discipline – an optional reviewer watches every turn and can intervene automatically.
• Flexibility – you can enable only the pieces you need (e.g., just the memory sync, or just the canvas) without bloating the AI’s toolset.

---

Who should consider using it

• Developers or knowledge workers who already use DSH and want their AI assistant to act as a long‑term partner rather than a stateless chatbot.
• Teams that need to share AI‑generated insights across multiple machines or collaborators.
• Anyone who runs many parallel projects and wants a single place to store tasks, logs, and reference files.
• Users who like the idea of AI‑driven “teams” where a main session coordinates specialised sub‑agents.

---

Limitations & notes (from the README)

• The plugin assumes you are running DSH version 260810 or newer (required for some notification APIs).
• Most features are disabled by default; you must turn them on in the settings UI.
• Memory writes are proposal‑only – the AI asks you to confirm before anything is persisted.
• Sync conflicts are resolved manually via a UI that lets you pick local, remote, or both versions.
• Updating the plugin is done through tagged Git releases; the UI will warn you and handle rollback if something goes wrong.

---

Bottom line

dsh‑memory‑evolve is a fairly comprehensive productivity extension for the DSH AI chat platform. It adds persistent, searchable memory, task management, multi‑agent orchestration, a visual asset board, and optional automated review—all aimed at turning a single AI chat into a collaborative, long‑term work partner.