What is OpenClaw?

OpenClaw is the open-source AI agent engine that MultiClaw runs on. It starts automatically when you open the desktop app, runs entirely on your machine, and stores everything as plain files. You don't install, configure, or manage it — MultiClaw handles all of that for you.

How OpenClaw and MultiClaw work together

MultiClaw and OpenClaw are two separate layers of the same product. OpenClaw is the execution layer: it runs your agents, manages conversation history, and stores your data. MultiClaw is the control layer: it provides the interface, connects to MultiClaw Cloud for team features, and manages the OpenClaw process.

Keeping them separate means the engine running your agents can be open source, independently auditable, and not tied to any single company's cloud.

The Gateway

When you open the desktop app, MultiClaw starts OpenClaw in the background. OpenClaw runs a local web server called the Gateway. The desktop app connects to the Gateway over WebSocket to send instructions and receive results.

Because the Gateway runs on your machine, agent execution doesn't require a cloud connection. Your agents can work even when MultiClaw Cloud is unreachable, as long as the desktop app is running.

When you see "Connected" in the status bar, the Gateway is up and the desktop app is communicating with it successfully.

Warning:

If the desktop app shows "Offline" after it was working, the Gateway may have stopped unexpectedly. Restart the desktop app to restart the Gateway.

What the open-source engine means for you

OpenClaw is open source on GitHub. This has three practical implications:

Transparency: Every line of code that processes your conversations and executes your agent tasks is publicly readable.
Portability: Your data is stored in open, human-readable formats, and the engine that reads them is open source. You're never locked in.
Independent auditability: Security researchers, developers, and enterprise teams can audit, contribute to, and verify the engine without relying on MultiClaw's word.

Where your data lives

OpenClaw stores everything as plain files on your machine.

Data type	Mac / Linux	Windows
Conversation history	`~/.openclaw/agents/` (`.jsonl` files)	`%USERPROFILE%\.openclaw\agents\` (`.jsonl` files)
Agent configuration	`~/.openclaw/openclaw.json`	`%USERPROFILE%\.openclaw\openclaw.json`
Skills	`~/.openclaw/workspace/skills/` (`SKILL.md` files)	`%USERPROFILE%\.openclaw\workspace\skills\` (`SKILL.md` files)
Scheduled jobs	`~/.openclaw/cron/jobs.json`	`%USERPROFILE%\.openclaw\cron\jobs.json`

Your configuration file

openclaw.json is OpenClaw's main configuration file. It stores agent definitions, runtime settings, and workspace preferences. MultiClaw reads and writes this file when you make changes through the desktop app.

You can open openclaw.json in any text editor to inspect or back up your settings. To change a setting not exposed in the desktop app, edit the file directly. The Gateway port is one example — restart the desktop app after any manual edits to apply the change.

Your data stays on your machine

All agent execution happens locally. Conversations, plans, and results are written to the files above, not sent to MultiClaw Cloud. Nothing from OpenClaw's runtime leaves your machine unless you explicitly share or sync it.

This matters if your work involves sensitive content: the agent runtime is local-only by design.

You don't need to manage it

MultiClaw installs OpenClaw automatically on first launch. When a new version is available, MultiClaw updates it the next time you open the desktop app — no action required. You can check the current OpenClaw version in Settings → About.

If you see openclaw in your system's process list or a log file, that's expected. It's the engine running in the background.

Tip:

Your data is stored as standard files, so you can back it up at any time. Copy ~/.openclaw/ to another machine to transfer your conversation history, agents, and skills.