OpenClaw Vector Memory

1. Course Content

Course Overview

OpenClaw has powerful memory capabilities that enable the robot to remember important information during conversations and intelligently retrieve and use these memories in subsequent interactions. This section will introduce OpenClaw's Vector Memory (Semantic Memory) system, which is an intelligent memory retrieval mechanism based on semantic understanding.

What is Vector Memory

Vector Memory is OpenClaw's built-in semantic search memory system, which implements intelligent memory retrieval through the following core technologies:

• Vector Embedding: Converts memory documents into high-dimensional vector representations, enabling computers to understand the semantic meaning of text

• Semantic Retrieval: Matches based on the semantic similarity of queries, not just keyword matching, can understand synonyms and related concepts

How It Works

1. Vectorization Processing

   • Converts content from MEMORY.md and memory/*.md into vector representations

   • Supports semantic understanding, not just keyword matching

2. Semantic Retrieval

   • Based on the semantic meaning of queries, not exact keywords

   • Returns the most relevant text fragments and their location information

Use Cases

Workflow

User Question → memory_search(Semantic Query) → Returns Relevant Fragments → memory_get(Read Specific Location) → Generate Response

Features

• Intelligent Matching: Understands synonyms and related concepts

• Automatic Citation: Results include file path and line number

• Complementary to File Memory: Vector memory for retrieval, file memory for persistent storage

2. Preparation

• First study the 【OpenClaw Skills Development】 chapter to master creating new skills

xxxxxxxxxx
ros2 launch yahboomcar_bringup mcp_service.launch.py use_camera:=true

• If you need voice replies, you need to add the use_tts parameter: ros2 launch yahboomcar_bringup mcp_service.launch.py use_camera:=true use_tts:=true, otherwise it can be omitted

3. Memory Feature Demo Case

• When performing inspection tasks in the future, follow the order of production workshop, material warehouse, and logistics center. Remember this inspection order and follow this process for future inspections.

• OpenClaw's memory content will be stored in files at the following path. You can view the memory content from the above case:

xxxxxxxxxx
cat $HOME/.openclaw/workspace/MEMORY.md

4. Retrieving Memory via Vector Search

• OpenClaw supports slicing and retrieving memory documents in vector mode, suitable for semantic retrieval of memories in large-scale memory documents

4.1 Prerequisites

• You need to register an account with an embeddings model provider. The following demonstration uses the openrouter platform as an example. For account registration and API application, refer to the tutorial 【AI Model Development—Register a Model Service Provider Account—3. openrouter Platform Account】

4.2 Configure embeddings Model API_KEY

• Set the model provider (openai interface protocol)

xxxxxxxxxx
openclaw config set agents.defaults.memorySearch.provider "openai"

• Set the baseUrl. Here using openrouter as an example. If you are using embeddings models from other providers, replace with the actual baseUrl

xxxxxxxxxx
openclaw config set agents.defaults.memorySearch.remote.baseUrl "https://openrouter.ai/api/v1"

• Set the embeddings model. Here using the "openai/text-embedding-3-small" model as an example. If you are using other embeddings models, replace with the actual name

xxxxxxxxxx
openclaw config set agents.defaults.memorySearch.model "openai/text-embedding-3-small"

• Set the API_KEY for the embeddings model

xxxxxxxxxx
openclaw config set agents.defaults.memorySearch.remote.apiKey "sk-or-v1-133949f96bec556343539a587f53d077xxxxxxxxxxxxxxxx"

• Restart the OpenClaw gateway to take effect

xxxxxxxxxx
openclaw gateway restart

Configuration Parameter Description:

• agents.defaults.memorySearch.provider: The model provider interface protocol for memory search. Setting it to "openai" means using an OpenAI-compatible interface

• agents.defaults.memorySearch.remote.baseUrl: The API service address of the embeddings model. Fill in the corresponding baseUrl according to the provider actually used

• agents.defaults.memorySearch.model: The name of the embeddings model, used to convert memory documents into vector representations. Different providers offer different embeddings models, so fill in according to the model actually used

• agents.defaults.memorySearch.remote.apiKey: The API key for the embeddings model, used for identity authentication

CLI Command Description:

• openclaw memory status --deep: Views the complete status of vector memory, including the sliced file paths and vector index information, used to confirm whether the configuration is in effect

• openclaw memory index --force: Forces a rebuild of the memory index. It needs to be executed manually once after initial configuration or memory document updates

• openclaw memory search "keyword": Tests semantic retrieval, returning the memory fragments most semantically relevant to the keyword along with scores. A higher score indicates higher semantic similarity (typically > 0.3 indicates good relevance)

4.3 Verify Vector Memory

• Check if vector memory is configured and active. The following output proves the configuration is successful, where Vector is the path of the sliced file

xxxxxxxxxx
openclaw memory status --deep

• After initial configuration, you can manually update the index once

xxxxxxxxxx
openclaw memory index --force

• Test whether memory retrieval can find the memory fragments from the demonstration case above

xxxxxxxxxx
openclaw memory search "inspection"

• Here 0.357 is the semantic relevance score. A higher score indicates higher semantic similarity. Results greater than 0.3 typically have good relevance