Back to directory

dsh-tdai-memory

In a curated listMaintenance: Active

scorp1o117/dsh-tdai-memory

TencentDB Agent Memory integration with conversation capture, structured memory extraction, automatic recall, and memory and conversation search tools.

View on GitHub
$ dsh

14

stars

2

forks

JavaScript

Language

MIT

License

2026-08-13

Created

2026-11-15

Last push

MIT-licensed DSH memory plugin: every conversation is captured into L0–L3 layers, recalled on later messages and searchable through two agent tools, with the data kept locally under `~/.memory-tencentdb/memory-tdai`.

DSH integration

Native runtime

Author-claimed

Safety audit

Unaudited

Last verified

2026-11-21

License

MIT

01What can it help you accomplish?

  • Give DSH conversations a layered long-term memory

    L0 raw conversation capture, L1 structured memories (facts, preferences, events), L2/L3 scene information and a user profile, all stored by default under `~/.memory-tencentdb/memory-tdai`

    DSH users who re-explain the same project background, preferences and history in every new session

  • Recall relevant memory and search it without leaving the conversation

    Memories recalled for the current message, plus `tdai_memory_search` over extracted memories and `tdai_conversation_search` over captured conversations

    Anyone whose DSH answers depend on decisions, preferences or facts established weeks earlier

  • Decide how much of the memory pipeline actually runs

    Three presets — auto memory (capture, extraction, auto recall and search tools all on), search existing memories only, or paused — plus per-toggle overrides for storage location and tuning parameters

    Users who want capture on but automatic recall off, or who only want to query memories they already have

02How to install into DeepSeek Harness

Prerequisites

  • DSH Web or DSH Desktop; the plugin accepts every host version from `0.2.0-rc.2` on, later prereleases included
  • Node.js `>=22.18`
  • An OpenAI-compatible LLM endpoint, model and key for extraction, plus an OpenAI-compatible embedding endpoint, model and dimension — the shipped default embedding address is a placeholder, not a working service

Installation steps

  1. 01

    Desktop: install the desktop `dsh` from 应用 → 管理 dsh 命令 first, then run `dsh plugin --profile desktop add dsh-tdai-memory`.

    $ dsh

  2. 02

    Web: run `dsh plugin --profile web add dsh-tdai-memory` instead.

    $ dsh plugin --profile web add dsh-tdai-memory

  3. 03

    Restart DSH, then open Plugins → Installed → dsh-tdai-memory.

  4. 04

    Alternative: install it through the Enhancement Suite instead of the CLI.

Verify the integration

  • After the restart the plugin is listed under Plugins → Installed → dsh-tdai-memory.
  • Ask the agent to search memory: both tools should answer, and the plugin log should show `[tdai-memory] tools registered`.

03DSH integration and capability boundaries

DSH integrationNative runtime

Native DSH plugin: install it with the `dsh plugin` CLI (`--profile desktop` or `--profile web`), restart DSH, and it appears under Plugins → Installed as dsh-tdai-memory, capturing conversations, extracting memory and registering two search tools

  • Layered capture and extraction (L0–L3)

    Every conversation turn captured while the plugin is enabled, plus the extraction model you configure→L0 raw conversations, L1 structured memories of facts, preferences and events, L2/L3 scene information and a user profile

    writes memories under `~/.memory-tencentdb/memory-tdai` by defaultsends extraction requests to the LLM endpoint you configureextraction runs in the background, so its latency and quality follow that model and service
  • Automatic recall per message

    The current message in an ongoing DSH conversation→Related memories retrieved for that message and put in front of the agent; UI language follows DSH

    sends embedding requests to the endpoint you configure on every recall
  • Memory and conversation search tools

    A query the agent writes, either for extracted memories or for raw conversations→`tdai_memory_search` results over the memory layers and `tdai_conversation_search` results over captured conversations

  • Memory modes and advanced overrides

    One of three presets (auto memory, search existing memories only, paused) or individual toggles→Capture, extraction, auto recall and search tools switched on or off, with storage location and tuning parameters adjustable

    the search-tool toggle applies immediately, other pipeline changes need a DSH restartthe settings form only writes edited fields, so keys you did not touch are preserved

04Who is it for? When not to use it?

Good for

  • DSH users who re-explain the same project background, preferences and history in every new session
  • Anyone whose DSH answers depend on decisions, preferences or facts established weeks earlier
  • Users who want capture on but automatic recall off, or who only want to query memories they already have

Not for

  • You have to supply the services: extraction needs an OpenAI-compatible LLM address, model and key, and recall needs an OpenAI-compatible embedding address, model and dimension. The default embedding address is only a configuration value, not a bundled service.
  • Captured conversations and memories stay on this machine, but extraction and embedding requests are sent to the services you configure — choose those endpoints yourself and back up the memory directory.
  • Nothing is guaranteed to be stored or recalled: writing, extraction and recall depend on the enabled state, the model connection and retrieval results, and the extraction model must emit stable structured output.

05Compatibility, maintenance and safety notes

  • You have to supply the services: extraction needs an OpenAI-compatible LLM address, model and key, and recall needs an OpenAI-compatible embedding address, model and dimension. The default embedding address is only a configuration value, not a bundled service.
  • Captured conversations and memories stay on this machine, but extraction and embedding requests are sent to the services you configure — choose those endpoints yourself and back up the memory directory.
  • Nothing is guaranteed to be stored or recalled: writing, extraction and recall depend on the enabled state, the model connection and retrieval results, and the extraction model must emit stable structured output.
  • It runs on DSH Web and Desktop from host version `0.2.0-rc.2` onwards, including later prereleases, and needs Node.js `>=22.18`; a fully local embedding backend additionally needs the optional `node-llama-cpp` build installed and approved in the target profile.
2026-08-132026-11-15v0.4.9

MIT · actively maintained (latest release v0.4.9, 2026-11-15; DSH port of the OpenClaw plugin)

06Frequently asked questions

How do I install it in DeepSeek Harness?

Desktop: install the desktop `dsh` from 应用 → 管理 dsh 命令, run `dsh plugin --profile desktop add dsh-tdai-memory`, then restart DSH and open Plugins → Installed → dsh-tdai-memory. On DSH Web the same command with `--profile web` applies. The Enhancement Suite can install it too.

What has to be configured before memory actually works?

An OpenAI-compatible LLM address, model and key for extraction, and an OpenAI-compatible embedding address, model and dimension for recall. The default embedding address is only a placeholder — you must run or point to a real embedding service. Then choose auto memory, save, and restart DSH so the pipeline picks up the configuration.

Which DSH versions and platforms are supported?

DSH Web and DSH Desktop. The plugin accepts every host version from `0.2.0-rc.2` onwards, later prereleases included, and needs Node.js `>=22.18`. A fully local embedding backend is the only case that needs the optional `node-llama-cpp` build, installed and approved in the target profile.

Where is my memory stored and what leaves the machine?

Captured conversations and memories live on this machine, under `~/.memory-tencentdb/memory-tdai` by default, and an existing store in that directory is reused. Extraction and embedding requests do go to the services you configured, so pick the endpoints deliberately and back up that directory.

How does it sit next to other DSH plugins?

It is a standalone DSH plugin installed through the `dsh plugin` CLI (or the Enhancement Suite) and loaded at DSH startup; it needs no companion plugin. It contributes two agent tools, `tdai_memory_search` and `tdai_conversation_search`, and follows DSH's own UI language.

08Data and sources

  • Author-claimedgithub.come426d833d404…

    将 TencentDB Agent Memory 的分层记忆系统接入 [DeepSeek Harness](https://github.com/deepseek-ai/deepseek-harness):捕获对话、提取结构化记忆,并在后续…

  • Author-claimedgithub.come426d833d404…

    按当前消息召回相关记忆;提供 `tdai_memory_search` 和 `tdai_conversation_search` 工具

  • Author-claimedgithub.come426d833d404…

    dsh plugin --profile desktop add dsh-tdai-memory

  • Author-claimedgithub.come426d833d404…

    安装后重启 DSH,打开 **插件 → 已安装 → dsh-tdai-memory**。

This page is generated from the project’s public documentation, repository metadata and a structured parse of DSH Plugins; last verified on 2026-11-21. Found an error? Submit a correction.

🏆

Best DeepSeek Harness Plugins

Twelve plugins worth installing first — picked from the whole catalog, across every category.

DSH Plugins is an independent community directory of DeepSeek Harness plugins. Not affiliated with or endorsed by DeepSeek. Third-party plugins are not security-audited — review the source before installing.

New DeepSeek Harness plugins, weekly. No spam.