@memtensor/memmy-memory-cli

Install#

Node.js 20 or later is required.

npm install -g @memtensor/memmy-memory-cli

During npm installation, the package runs node scripts/postinstall.js. The npm package is only a Node launcher shell. It downloads the platform-specific memmy-memory binary archive for the current package version and extracts it to the package bin/ directory.

Supported targets:

• darwin-arm64
• darwin-x64
• linux-arm64
• linux-x64
• windows-arm64
• windows-x64

Environment variables:

• MEMMY_MEMORY_INSTALL_SKIP_DOWNLOAD=1 skips the binary download.
• MEMMY_MEMORY_BINARY_URL=<url> uses a custom archive URL.

Default binary URL:

https://github.com/MemTensor/memmy-agent/releases/download/memory-v{version}/memmy-memory-{version}-{target}.tar.gz

For example, a macOS arm64 archive name is:

memmy-memory-{version}-darwin-arm64.tar.gz

If download is skipped, running memmy-memory will fail until the binary exists at bin/memmy-memory or bin/memmy-memory.exe inside the installed package.

Setup#

Initialize Memory CLI config:

memmy-memory init

init writes the Memory endpoint and optional local SQLite path to the Memmy config file. Install the standalone service without changing an Agent with:

memmy-memory install --service-only

Install the service and adapters for detected Agents, or upgrade the active runtime and its installed adapters, with:

memmy-memory install
memmy-memory install --agents openclaw,hermes
memmy-memory upgrade
memmy-memory upgrade --version 2.1.0

By default, init installs agent-side files for each supported agent root it finds and skips agents that are not installed. Use --agent to require and install only selected agents, or --skip-agent-skills when only the Memory configuration should be written:

memmy-memory init --agent codex
memmy-memory init --agent codex,cursor,claude
memmy-memory init --skip-agent-skills

Supported agents:

• codex
• cursor
• claude
• opencode
• openclaw
• hermes
• dsh
• workbuddy
• pi
• qwenwork

Commands#

memmy-memory --help
memmy-memory --version
memmy-memory init
memmy-memory health
memmy-memory reload-config
memmy-memory serve
memmy-memory session open --source <agent-source>
memmy-memory turn start --source <agent-source> --session-id <sessionId> --query "<query>"
memmy-memory turn complete <turnId> --source <agent-source> --session-id <sessionId> --query "<query>" --answer "<answer>"
memmy-memory search "<query>" --source <agent-source>
memmy-memory search "<query>" --source <agent-source> --verbose
memmy-memory add "<content>" --source <agent-source>
memmy-memory get <id>
memmy-memory get <id> --verbose
memmy-memory delete <id>
memmy-memory raw GET /panel/overview

memmy-memory install downloads a verified, versioned Memory runtime, registers the user-level background service, starts it, and verifies /api/v1/health. The development-only --source-path option retains the source-tree symlink flow.

memmy-memory get <id> prints compact agent-readable memory content by default. Use --verbose when debugging the full JSON detail payload.

By default the CLI talks to http://127.0.0.1:18960. Use --url <url>, --token <token>, or --config <path> to target a specific Memory HTTP service or config file. Use --user-id <id> or --user_id <id> when a single command must target a specific Memory namespace user. If omitted, the CLI uses the active Memory profile's user ID and then the application user ID from the configured Memmy configuration. Configure the application default with memmy config set app.userId <user_id>. Use --source <agent-source> to identify the calling agent/source, such as codex, cursor, or openclaw.