Skip to main content
Glama
MnemosyneOS

mnemosyne

Official

dsh-ai-memory

Mnemosyne OS 8.1.0 memory for DeepSeek Harness — auto-register, auto-remember, auto-recall.

One command, one click in the browser, and your DSH sessions remember across restarts. 52 MCP tools (write, recall, knowledge graph, memory formation layer) with no glue code.


Install (30 seconds)

dsh plugin add github:MnemosyneOS/dsh-ai-memory
dsh-ai-memory-login

Restart dsh web. That's it.

Requires Node 18+ and pnpm on PATH (dsh plugin calls pnpm under the hood).

Related MCP server: aura-memory

First authorization (Device Flow)

dsh-ai-memory-login prints a short link:

请在浏览器打开下面这个地址完成授权:

    https://ai-memory.net/api/device/confirm?code=ABCD2345

Open it, sign in, click 授权. The command then:

  1. receives a mn- key bound to your namespace,

  2. writes exactly one line (Authorization: Bearer mn-…) into your profile's cordis.patch.yml,

  3. backs the file up first as .bak-dsh-ai-memory,

  4. never touches any other line — your comments and other MCP servers stay byte-identical.

The key is shown as mn-Abcd****wxyz; the full value is never logged.

What it gives you

The plugin registers one MCP server (mnemosyne, streamable HTTP at https://ai-memory.net/v1/mcp), which exposes the full Mnemosyne OS tool surface. Commonly used:

Tool

What it does

retain / retain_batch

Write memories (content / type / tags / confidence / importance)

recall

Five-channel recall with token budget and compression

forget

Soft forget: confidence → 0, marked deleted, still auditable

capsule / expand

Compress a memory to pointer + fact atoms; expand back and verify its hash

recall_catalog / expand_memory

Progressive disclosure: titles first, full text on demand

graph_query / temporal_query

Walk the knowledge graph; return version chains by time

memory_inject / profile_card

Build injectable context within a token budget

session_summarize / topic_summarize

Session and project-level summaries

conflict_resolve

Six conflict classes, four resolution operations

code_change_record / code_archaeology

Record intent, constraints and rejected approaches; archaeology on past changes

Then in any session: "记住:我叫张三,乳糖不耐" — and in a brand-new session, "我喝咖啡要注意什么" gets the right answer.

Privacy & data

  • Where data lives. In your Mnemosyne namespace on ai-memory.net, isolated per account.

  • Isolation is server-side. The namespace is decided by the server from your session (default for administrators, u_<sha256(email)[:16]> for members). A namespace supplied by a client is ignored. Another user's key cannot reach your memories.

  • Keys. Only SHA-256 digests are stored; the plaintext exists once, at issuance. Revoke or regenerate keys in the ai-memory.net console.

  • Delete / opt out. dsh plugin remove dsh-ai-memory removes the plugin; run dsh-ai-memory-login --remove to strip the profile entry, and delete the namespace from the console to erase stored memories. The DSH-side config file is yours: the login tool only ever edits one line, keeps a .bak-dsh-ai-memory backup, and rolls back if its own check fails.

  • No telemetry. The plugin talks to ai-memory.net only (device flow + MCP). Nothing else.

Uninstall

dsh-ai-memory-login --remove          # drop the MCP entry from your profile
dsh plugin remove dsh-ai-memory       # remove the package

Restart dsh web.

Options

--profile <name>   target DSH profile (default: web)
--file <path>      point at a cordis.patch.yml directly
--base-url <url>   override the service base URL (default https://ai-memory.net)
--no-browser       do not try to open the browser automatically
--remove           strip the entry this plugin wrote
-h, --help         help

Exit codes: 0 success / already configured · 1 failure · 2 cancelled or timed out.

Development

node test/login.test.mjs   # 26 checks: protocol + idempotent profile writes

License

MIT © MnemosyneOS

Maintenance

ActivityMaintained
ResponsivenessNo issues

Related MCP Connectors

Related MCP Servers

  • F
    license
    B
    quality
    D
    maintenance
    Enables persistent memory for AI systems by providing tools for episodic, semantic, and procedural data storage through a vector-and-graph-enhanced database. It allows models to maintain long-term continuity using similarity search, thematic clustering, and identity tracking.
    24
    1
    -
  • A
    license
    Not graded
    quality
    A
    maintenance
    A persistent, project-scoped memory layer for AI coding agents, providing 39 tools for capturing, searching, deduplicating, relating, and maintaining memory records across long-running coding sessions.
    0
    MIT
  • F
    license
    A
    quality
    B
    maintenance
    Shared fleet memory for AI agents — what one agent learns, the whole fleet knows. Nightly self-correction: stale facts rewritten in place, duplicates merged. Recall pushed into every prompt in supported coding agents. Self-hosted.
    11
    3
    -