TencentDB Agent Memory: Enhancing AI Agents with Layered Long-Term Memory

Introduction

TencentDB Agent Memory is an advanced solution designed to equip AI agents with robust, fully local long-term memory capabilities. It introduces a novel 4-tier progressive pipeline, ensuring zero external API dependencies for memory management. This project addresses critical challenges in AI agent performance, particularly in reducing token consumption and enhancing task success rates over long-horizon sessions. By moving beyond flat storage, TencentDB Agent Memory enables agents to remember, learn, and reason more effectively.

Installation

The project supports integration with OpenClaw and Hermes agents.

For OpenClaw:

Install the plugin:

openclaw plugins install @tencentdb-agent-memory/memory-tencentdb
openclaw gateway restart

Enable zero-config:
Add the following to ~/.openclaw/openclaw.json:

// ~/.openclaw/openclaw.json
{
  "memory-tencentdb": {
    "enabled": true
  }
}

Enable short-term compression (optional, requires version ? 0.3.4):
First, update ~/.openclaw/openclaw.json to enable offloading:

{
  "memory-tencentdb": {
    "config": {
      "offload": {
        "enabled": true
      }
    }
  }
}

Then, register the slot in your plugin config:

{
  "plugins": {
    "slots": {
      "contextEngine": "memory-tencentdb"
    }
  }
}

Finally, apply the runtime patch:

bash scripts/openclaw-after-tool-call-messages.patch.sh

For Hermes:

There are two main ways to integrate with Hermes:

Docker (greenfield, requires version ? 0.3.4):
Navigate to the docker/opensource directory in the cloned repository.
Build the Docker image:

docker build -f Dockerfile.hermes -t hermes-memory .

Run the container, replacing your-api-key with your actual key:

docker run -d \
  --name hermes-memory \
  --restart unless-stopped \
  -p 8420:8420 \
  -e MODEL_API_KEY="your-api-key" \
  -e MODEL_BASE_URL="https://api.lkeap.cloud.tencent.com/v1" \
  -e MODEL_NAME="deepseek-v3.2" \
  -e MODEL_PROVIDER="custom" \
  -v hermes_data:/opt/data \
  hermes-memory

Verify the Gateway:

curl http://localhost:8420/health

Enter the Hermes interactive shell:

docker exec -it hermes-memory hermes

Attach to Existing Hermes (No Docker):

Download the plugin package:

mkdir -p ~/.memory-tencentdb
TEMP_DIR=$(mktemp -d)
cd "$TEMP_DIR"
npm init -y --silent
npm install @tencentdb-agent-memory/memory-tencentdb@latest --omit=dev
cp -r node_modules/@tencentdb-agent-memory/memory-tencentdb \
              ~/.memory-tencentdb/tdai-memory-openclaw-plugin
rm -rf "$TEMP_DIR"

Install Gateway dependencies:

cd ~/.memory-tencentdb/tdai-memory-openclaw-plugin
npm install --omit=dev
npm install tsx

Link to the Hermes plugin directory:

rm -rf ~/.hermes/hermes-agent/plugins/memory/memory_tencentdb
ln -sf ~/.memory-tencentdb/tdai-memory-openclaw-plugin/hermes-plugin/memory/memory_tencentdb \
               ~/.hermes/hermes-agent/plugins/memory/memory_tencentdb

Declare the provider in ~/.hermes/config.yaml:
```
memory:
  provider: memory_tencentdb
```

Configure Gateway environment variables in ~/.hermes/.env:

MEMORY_TENCENTDB_GATEWAY_CMD="sh -c 'cd ~/.memory-tencentdb/tdai-memory-openclaw-plugin && exec npx tsx src/gateway/server.ts'"
MEMORY_TENCENTDB_GATEWAY_HOST="127.0.0.1"
MEMORY_TENCENTDB_GATEWAY_PORT="8420"
TDAI_LLM_API_KEY="sk-your-api-key-here"
TDAI_LLM_BASE_URL="https://api.openai.com/v1"
TDAI_LLM_MODEL="gpt-4o"

Start the Gateway (auto-discovery or manual run):

cd ~/.memory-tencentdb/tdai-memory-openclaw-plugin
npx tsx src/gateway/server.ts

Verify:
```
curl http://127.0.0.1:8420/health
```

Examples

The project's README includes video demonstrations showcasing TencentDB Agent Memory in action with both OpenClaw and Hermes. These demos visually illustrate how the memory system enhances agent performance, particularly in handling complex, long-horizon tasks and maintaining context.

Why Use TencentDB Agent Memory

TencentDB Agent Memory offers significant advantages for AI agents, moving beyond traditional flat memory systems. Its core technologies, memory layering and symbolic memory, provide a robust framework for improved agent performance.

Reduced Token Usage and Improved Success Rates: By offloading heavy tool logs and condensing information into compact Mermaid symbols, the system drastically cuts token usage, achieving up to a 61.38% reduction. This leads to a substantial increase in task success rates, with benchmarks showing a 51.52% relative improvement in some scenarios.
Layered Long-Term Memory: Instead of fragmented vector stores, it distills conversations into a semantic pyramid, from raw dialogue (L0) to atomic facts (L1), scene blocks (L2), and user personas (L3). This hierarchical structure ensures that agents remember what's important and can drill down for details when necessary.
Symbolic Short-Term Memory: Task state transitions are encoded using high-density Mermaid syntax. This allows agents to reason over a concise symbol graph, while full logs are offloaded, maintaining full traceability via node_id for detail retrieval.
White-Box Debuggability: Unlike opaque memory systems, TencentDB Agent Memory keeps key intermediates as readable files. Scenario blocks are Markdown, personas are in persona.md, and task canvases are Mermaid, making debugging a transparent process from high-level abstractions to ground-truth evidence.
Production-Ready Engineering: The project is designed for real-world applications, offering an OpenClaw plugin, Hermes Gateway adapter, local SQLite backend with sqlite-vec, and hybrid retrieval (BM25 + vector + RRF) for both keyword and semantic recall.