dsh-memory-recall

A DeepSeek Harness plugin that gives the model a memory, in the style of Hermes Agent: what it learns about you and your projects carries into every new chat, and it can look back through your earlier chats — only after you allow it.

Everything stays on your machine. Two small text files, dsh’s own session index, no embeddings, no service, no telemetry.

View on GitHub

Install

dsh plugin --profile web add dsh-memory-recall
dsh --profile web

Or install from GitHub or a local folder instead:

dsh plugin --profile web add github:adithyanraj03/dsh-memory-recall
dsh plugin --profile web add /path/to/dsh-memory-recall

Then restart dsh if it was already running and hard-refresh the browser (Ctrl+Shift+R / Cmd+Shift+R) — a plain refresh can serve a cached client bundle. 🧠 Memory appears at the bottom of the sidebar.

In the desktop app

The desktop dsh installs from its own Plugins settings: click Add plugin, paste dsh-memory-recall (or the GitHub address), and click Install. The dialog confirms with the Installed state. Installed plugins do not update automatically yet — to upgrade, uninstall and install again.

Turn on dsh’s full-text search (recommended)

session_search uses dsh’s own session index, which dsh ships switched off. Add this to your profile’s cordis.patch.yml (~/.dsh/profiles/web/cordis.patch.yml):

- id: session-query-sqlite
  config:
    path: '/absolute/path/to/.dsh/session-search.db'
    openAt: first-search

first-search opens the index only when something first searches; a file path keeps it across restarts. Without it, session_search still works — a literal scan of your 150 most recent chats, slower and unranked, and it says so in its card.

What it does

A memory the model keeps

The model maintains two short files: a user profile (who you are, how you like to work) and its own notes (environment facts, conventions, lessons). Every chat gets both as a context message logged in the chat as 🧠 Memory loaded — at the start, and again after a change or a compaction.

Just talk. The model saves what will matter later, and each save appears in the chat with its own budget:

Saves in the chat: an added note and an updated profile entry, each with its budget

The five states a save can end in — saving, saved, updated, full, and refused by the scan:

The chip's five states

The 🧠 button in the message box is lit when this chat has memory, dim when it has none, amber when the memory changed since — click it to see what this chat is running on.

A Memory panel

The Memory panel with the model's notes, a budget bar and inline edit

Everything kept, in two tabs — Notes and Profile — each with a ring and a bar showing how much of the budget is used. Edit or delete any entry in place (hover shows the actions), add your own, and search once a tab holds a few. ⚙ Display settings toggle the 🧠 button and the @ rows per browser — and carry the budget sliders: drag the notes and profile character budgets (≈ tokens shown), saved with the memory files and applied from the chat’s next delivery; reset returns to the limits in cordis.patch.yml. With the sidebar collapsed it becomes a single icon, like Settings.

The Profile tab in the light theme

Looking back, with permission

Ask “what did we decide last time?” and the model calls session_search. dsh’s own approval prompt takes over the composer first, saying what it wants to read, with Allow and Deny. The card it returns holds the chats’ actual messages — nothing summarised, nothing invented — each with a date, the matching line and a link to open it:

A recall card: three earlier chats found for "deploy rollback", with their matching lines

One Allow covers that reply. The search is usually followed by one close read of the best hit; asking twice per question would be noise. The permission ends when the reply ends, and the next question asks again.
Deny means nothing is read, and the model is told not to re-ask in that reply.
Saves never ask — they always show in the chat and can be undone from the panel.

Chats and workspaces in the `@` menu

Type @ in the message box. Under dsh’s own files and folders, your chats are grouped by workspace, under their real titles. Picking a chat attaches its read-only snapshot to the message — no search, no prompt, because you chose it. Picking a workspace sends [workspace "Name" — /path], and session_search then searches only that workspace’s chats, still asking first.

Tools

tool	what it does
`memory`	the two stores — read, add, replace, remove, batch — with the budget checked on the final result
`session_search`	earlier chats behind dsh’s approval prompt: best match, a chat’s messages, or a listing

Subagent chats get neither — the context message and the tools go to top-level chats only.

Configuration

On the plugin’s row in your profile’s cordis.patch.yml — all optional:

- id: dsh-memory-recall
  config:
    dir: '/path/to/store'     # where USER.md and MEMORY.md live (default ~/.dsh/dsh-memory-recall)
    limits:                   # character budgets (Hermes's defaults); the panel's sliders override them once moved, until reset
      memory: 2200
      user: 1375
    recall: false             # drop session_search, keep memory
    askBeforeRecall: false    # read earlier chats without asking

An override replaces the row’s whole config, so re-write every key you set. Restart dsh after a change.

Privacy

No telemetry, no network. Memory is plain text in ~/.dsh/dsh-memory-recall/ (USER.md, MEMORY.md, and settings.json once a slider moves), entries separated by a § line — Hermes’s format, readable and editable by hand. Writes the model makes are scanned with Hermes Agent’s threat patterns and refused with the reason; manual edits and panel edits are never scanned.

Requirements

dsh 0.1.5 or later, with a web profile — from plugin 0.2.1 also dsh 0.1.7+ and 0.2.x
Node 22+
No runtime dependencies — the client bundle requires only react

The design — two bounded stores, a frozen per-chat snapshot, batch-against-final-budget, session_search’s shapes and the threat patterns — comes from Hermes Agent by Nous Research (MIT). The brain icons are Lucide’s (ISC) and Phosphor’s (MIT).

Licensed MIT. Issues and pull requests at github.com/adithyanraj03/dsh-memory-recall.