OpenClaw

There are three ways to connect OpenClaw: paste one instruction and let OpenClaw do the rest, install the native plugin yourself with the hm CLI, or add HyperMemory as an MCP server.

Install with one instruction (recommended)

Paste this into OpenClaw. It runs the commands for you:

Install the HyperMemory CLI with "pipx install hypermemory-cli" (first make sure Python
3.10 or newer and Git are installed on this computer and install them if they are
missing), then run "hm login" so the HyperMemory OAuth sign-in page opens in my browser
and wait for me to approve it, then run "hm setup openclaw" to download and install the
latest HyperMemory skill and the native HyperMemory plugin into OpenClaw, and finally
run "hm doctor" to check that everything works.

OpenClaw installs the CLI, opens HyperMemory’s sign-in page in your browser, installs and enables the hypermemory-cli plugin, and checks the setup.

Option 2: Native plugin, by hand

Install the CLI and sign in

pipx install hypermemory-cli
hm login

Install the plugin

hm setup openclaw

This installs the hypermemory-cli OpenClaw plugin and enables it in ~/.openclaw/openclaw.json.

The plugin uses two OpenClaw hooks:

Hook	What it does
`before_prompt_build`	Recalls from HyperMemory (`hm recall`) and adds the results to the prompt.
`agent_end`	Writes a timeline entry and reports the run’s token usage.

Remove it with hm uninstall openclaw.

Option 3: MCP server

Add HyperMemory to your OpenClaw MCP configuration:

{
  "mcpServers": {
    "hypermemory": {
      "url": "https://api.hypermemory.io/mcp"
    }
  }
}

OpenClaw runs the OAuth sign-in when it first connects. To use an API key instead, add "headers": { "Authorization": "Bearer hm_YOUR_KEY" }.

With MCP, give the agent the HyperMemory protocol too: ask it to fetch the skill with hm_skill (variant: "general") and follow it.

A typical agent loop

Start of a session — hm_get_overview, then hm_recall with the user’s words.
New information — hm_recall to check for an existing node, then hm_store or hm_update.
Dense text — hm_ingest, then clean up orphans.
End of a turn — hm_timeline_write for milestones and hm_tokens to report usage.