All Products
Search

This Product

    Cloud Monitor:FS quick start

    Document Center

    Cloud Monitor:FS quick start

    Last Updated:Mar 28, 2026

    Map the AgentLoop Memory Store service to a local virtual file system. This lets an agent autonomously explore and manage memories using standard shell commands such as ls, cat, echo, and rm, enabling context retrieval through the Pull paradigm.

    Prerequisites

    • An Alibaba Cloud account with the AgentLoop Memory Store service activated.

    • A workspace and a Memory Store created in the console.

    • An AccessKey ID and AccessKey Secret.

    • A Linux or macOS system.

    Procedure

    Step 1: Install the plugin

    Method 1: Automated installation

    Provide the contents of the INSTALL_PROMPT.md file to your agent, such as Cursor Agent or Claude Code. The agent will automatically perform the following tasks:

    • Install the underlying FUSE dependencies.

    • Download the Python package.

    • Configure environment variables.

    • Mount the file system in the background.

    You can even watch the agent troubleshoot errors, check logs, and resolve dependencies until it successfully mounts the file system.

    Method 2: Manual installation

    If you prefer to install the plugin manually, follow these steps:

    Step 1: Install FUSE dependencies

    Linux (Ubuntu/Debian):

    sudo apt-get install fuse

    Linux (CentOS/RHEL):

    sudo yum install fuse

    macOS:

    brew install macfuse
    Note

    On macOS, you must allow the macFUSE kernel extension in System Settings > Privacy & Security.

    Step 2: Install Python package

    pip install agentloop-memory-fs

    Step 3: Configure environment variables

    export ALIBABA_CLOUD_ACCESS_KEY_ID=your_access_key_id
export ALIBABA_CLOUD_ACCESS_KEY_SECRET=your_access_key_secret
export CMS_ENDPOINT=cms.cn-hangzhou.aliyuncs.com
export CMS_WORKSPACE=your_workspace_name
export CMS_MEMORY_STORE=your_memory_store_name

    Step 4: Mount the file system

    # Create a mount point.
mkdir -p /tmp/agentloop_memory

# Start the file system. This command runs it as a background process.
agentloop-memory-fs /tmp/agentloop_memory &

    Step 2: Integrate with AI agent

    System Prompt integration is key to implementing the Pull paradigm—enabling the agent to autonomously decide when to retrieve memories, rather than passively receiving them.

    Configure System Prompt

    Add the following instructions to your agent's System Prompt:

    Agent

    Configuration file location

    Qwen CLI

    ~/.qwen/QWEN.md

    Claude Code

    ~/.claude/CLAUDE.md or CLAUDE.md in the project root

    Cursor

    .cursor/rules or the project-level .cursorrules

    # Memory
You have persistent, cross-session memory, mounted at `/tmp/agentloop_memory/store/`.
- Search: `cat search/{query}.txt` — semantic search, where `query` is a natural language keyword.
- Write: `echo "内容" > _add.txt` — This operation is processed asynchronously.
- If you are unsure about a path, use `cat _help.txt`.
    Why is this enough? The agent already has shell command capabilities. You only need to tell the agent the memory's location and usage. It can then decide when and what to retrieve.

    Example workflow

    After configuration, the agent performs the following actions autonomously:

    1. Identifies an information gap → The agent recognizes it needs to understand the user's background.

    2. Proactively searches → cat /tmp/agentloop_memory/store/search/user_tech_stack.txt.

    3. Retrieves results → FUSE returns the top-N most relevant memories.

    4. Generates a response → The agent uses the retrieved memories to provide a precise response.

    5. Persists new information → echo "User decided to adopt a microservices architecture" > /tmp/agentloop_memory/store/_add.txt.

    Step 3: Verify

    # 1. Add a memory.
echo "我喜欢用 Vim 编辑器" > /tmp/agentloop_memory/store/_add.txt

# 2. Wait for asynchronous processing.
sleep 3

# 3. Search for the memory.
cat "/tmp/agentloop_memory/store/search/editor_preference.txt"

    Expected output:

    [
  {
    "id": "019c269b-...",
    "memory": "The user prefers using Vim editor.",
    "score": 0.89
  }
]

    Unmount file system

    fusermount -u /tmp/agentloop_memory  # Linux
umount /tmp/agentloop_memory         # macOS

    Core operations

    Intent

    Shell command

    API

    Browse directory structure

    ls /tmp/agentloop_memory/store/

    -

    View help

    cat /tmp/agentloop_memory/store/_help.txt

    -

    Add memory

    echo "用户偏好深色模式" > /tmp/agentloop_memory/store/_add.txt

    POST /memories

    Semantic search

    cat /tmp/agentloop_memory/store/search/ui_preference.txt

    GET /search?q=UI preference

    View specific memory

    cat /tmp/agentloop_memory/store/memories/{id}.json

    GET /memories/{id}

    Delete memory

    rm /tmp/agentloop_memory/store/memories/{id}.json

    DELETE /memories/{id}

    View all user memories

    cat /tmp/agentloop_memory/store/users/{user_id}/_all.txt

    GET /users/{user_id}/memories