Local-first, multi-agent persistent memory for AI assistants. Use it to install, operate, connect, diagnose, or update Memoria on macOS.

Install

openclaw skills install @nieto42/openclaw-memoria

Memoria

Memoria keeps persistent, local memory for AI assistants. It provides a local daemon, a web UI, an MCP server, a CLI, and a thin OpenClaw adapter for automatic recall and capture.

Use this skill when

  • Installing or updating Memoria on macOS.
  • Connecting an agent to its own persistent memory.
  • Inspecting memory health, the local daemon, pairing, import, sharing, or local synchronization.
  • Diagnosing why recall or capture is unavailable.

Install

Memoria is a public beta. It requires Node.js 20+; Node.js 22 LTS is recommended. On macOS, Apple Command Line Tools are also required.

sh
curl -fsSL https://raw.githubusercontent.com/Primo-Studio/openclaw-memoria/memoria-v1/scripts/install-memoria.sh | sh

The installer clones the memoria-v1 branch, installs dependencies, builds the project, initializes and starts the local daemon, enables login autostart on macOS, and opens the local web UI.

Everyday operations

sh
memoria ui       # open the local interface
memoria doctor   # inspect daemon, storage, and provider health
memoria agents   # list connected agents
memoria update   # update the local checkout and restart the service

Use the UI's Agents screen to detect, pair, and import agents. For terminal pairing, run memoria pair <agent-type> and use the one-time command it returns.

OpenClaw integration

The OpenClaw adapter is intentionally thin: it calls the Memoria daemon on localhost for recall before a prompt and capture after a turn. It needs a paired instance token. When enabling automatic capture, make sure the OpenClaw plugin entry allows conversation-hook access; the installer/onboarding configures this during connection.

If Memoria is unavailable, agents should continue normally without memory rather than failing the request.

Privacy and safety

  • Data is local by default in Memoria-managed SQLite storage.
  • Remote LLM providers are optional; configure one only with the user's approval.
  • Do not expose pairing codes, instance tokens, admin tokens, or local UI URLs containing tokens.
  • Use the review and sharing controls before making memories available across agents or machines.
  • Deleting memories is permanent; confirm the target before running a forget/delete operation.