Scorp1o117/dsh-tdai-memory
30Last commit Aug 16, 2026
dsh-tdai-memory DSH plugin
dsh-tdai-memory is a DSH plugin that brings a four-layer memory system (L0 raw conversation, L1 structured memory, L2 scenes, L3 persona) to any DSH agent. It captures every turn, extracts facts/preferences/events via an LLM, and injects relevant memories into the model's context automatically. The data directory is compatible with the original TencentDB memory, allowing seamless carryover of existing memories.
How to install the dsh-tdai-memory DSH plugin
dsh plugin --profile web add dsh-tdai-memoryCopying does not run this command. Review the repository and version before installing the dsh-tdai-memory DSH plugin.
dsh-tdai-memory DSH plugin data source
dsh-tdai-memory DSH plugin snapshot date: Aug 16, 2026
discovered
What the dsh-tdai-memory DSH plugin can do
- L0 conversation capture: writes every turn to JSONL + SQLite + FTS + vectors
- L1 structured memory extraction: background pipeline uses LLM to extract facts, preferences, and events
- L2 scenes and L3 persona generation: scheduled pipeline for scene blocks and user profiles
- Automatic recall injection: relevant memories and user profile are retrieved and injected into prompt assembly
- Tools: `tdai_memory_search` (L1 structured search) and `tdai_conversation_search` (L0 raw-text search)
Where the dsh-tdai-memory DSH plugin fits
- Enable the model to remember user preferences and past conversations across sessions
- Automatically extract and store factual information from conversations for later use
- Create persistent user profiles that adapt over time without manual configuration
- Search through conversation history or structured memories using dedicated tools
- Migrate existing TencentDB Agent Memory data to DSH without re-collection
Who the dsh-tdai-memory DSH plugin is for
- DSH users who need long-term memory capabilities for their agents
- Developers porting from the TencentDB Agent Memory ecosystem to DSH
dsh-tdai-memory DSH plugin limitations
- Requires external LLM and embedding API endpoints (configured in settings.yaml)
- Extraction model (mimo-v2.5) takes 20-30 seconds per call; deepseek-v4-flash produces invalid JSON
- Dedup feature is unstable and disabled by default; only enable with a more reliable model
- Configuration changes require a restart of `dsh web` (TdaiCore built at startup)
dsh-tdai-memory DSH plugin: from the repository README
Quoted from the Scorp1o117/dsh-tdai-memory README, the upstream source of the dsh-tdai-memory DSH plugin. Copyright remains with the original authors.
**GitHub**: [Scorp1o117/dsh-tdai-memory](https://github.com/Scorp1o117/dsh-tdai-memory) · **npm**: [dsh-tdai-memory](https://www.npmjs.com/package/dsh-tdai-memory) [](https://github.com/Scorp1o117/dsh-enhancement-suite) [](https://www.npmjs.com/package/dsh-enhancement-suite) Part of the [DeepSeek Harness Enhancement Suite](https://github.com/Scorp1o117/dsh-enhancement-suite) — Vision · Soul/Persona · Long-term Memory · Plugin Marketplace. A port of **TencentDB Agent Memory** (Tencent Cloud's open-source four-layer memory system, originally an OpenClaw plugin) into DeepSeek Harness. ## Features - **L0 conversation capture**: every turn (turn end, request boundary) is written to raw conversation storage (JSONL + SQLite + FTS + vectors) - **L1 structured memory**: a background pipeline uses an LLM to extract facts / preferences / events (persona / episodic / instruction) from conversations, stored in `records/` + SQLite + FTS + vectors - **L2 scenes / L3 persona**: scene blocks and user profile generation (pipeline-scheduled) - *
Read the full READMERepository license: MIT
dsh-tdai-memory DSH plugin questions
How do I install dsh-tdai-memory?
Run `dsh plugin --profile web add dsh-tdai-memory` in your terminal. This installs the plugin and mounts it automatically. Then restart `dsh web` for the settings section to appear. If the settings page doesn't show the memory section, restart once more.
What API keys do I need to configure?
You need an LLM API key for extraction (e.g., mimo-v2.5) and an embedding API key for vector storage. Set them in `$DSH_HOME/settings.yaml` under `tdai-memory:` section. You can also configure them via the Web UI Settings → 记忆 page after the plugin is active.
Where is the memory data stored?
The data is stored in the directory `~/.memory-tencentdb/memory-tdai` by default. This is the same location used by the original TencentDB Agent Memory, so existing memories are automatically carried over.
Does the plugin support Chinese?
Yes, the plugin is language-agnostic. It captures conversations in any language and extracts facts/preferences from them. The structured memory and search tools work with the content as provided.
How do I upgrade the plugin when new upstream code is released?
After pulling the latest code from the repository, run `npx tsc -p dsh-tsconfig.json` in the plugin project directory to compile the TypeScript source. The output goes to `dist-dsh/`. Then restart DSH to use the updated version.