Guides on OpenClaw Deploy

Agents & Sessions

Mon, 01 Jan 0001 00:00:00 +0000

Agents

OpenClaw supports multiple isolated agents, each with their own workspace and model. Define them in openclaw.json under agents.list.

Adding an Agent

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16


{
 "agents": {
 "list": [
 {
 "id": "main",
 "workspace": "~/.openclaw/workspace",
 "default": true
 },
 {
 "id": "researcher",
 "workspace": "~/.openclaw/workspace-researcher",
 "model": "anthropic/claude-haiku-4-5"
 }
 ]
 }
}

Then deploy:

Sandbox Configuration

Mon, 01 Jan 0001 00:00:00 +0000

Sandbox Configuration Guide

This guide explains how to enable agent sandboxing in OpenClaw. Sandboxing isolates tool execution from the host system, providing an extra layer of security.

Note: This guide is for the openclaw-deploy repository. For the full OpenClaw sandboxing documentation, see Gateway Sandboxing.

Why Enable Sandboxing?

Without sandboxing, agent tool execution (file operations, shell commands) runs directly on your VPS. If the agent is compromised or makes a mistake, the blast radius is your entire server.

Lossless Context

Mon, 01 Jan 0001 00:00:00 +0000

lossless-claw — Lossless Context Management

What

lossless-claw is an OpenClaw plugin that replaces the built-in sliding-window compaction with a DAG-based summarization system. Instead of truncating older messages when context fills up, it:

Persists every message in a SQLite database
Summarizes chunks of older messages into compressed nodes
Condenses summaries into higher-level nodes as they accumulate (forming a DAG)
Assembles context each turn by combining summaries + recent raw messages
Provides agent tools (lcm_grep, lcm_describe, lcm_expand) so the agent can search and recall details from compacted history

Why we install it

The problem: Without LCM, OpenClaw’s default compaction simply drops older messages when context fills up. In long conversations (multi-hour sessions, daily check-ins), context gets lost — the agent forgets decisions made hours or days ago.

Skills & Tools

Mon, 01 Jan 0001 00:00:00 +0000

Skills

OpenClaw is extended via skills — slash commands, hooks, templates, and tool definitions. docker/skills-manifest.txt lists which skills are auto-installed at container startup via clawhub install.

Adding a ClawHub Skill

ClawHub is the community skill registry.

Find the skill at clawhub.ai (e.g. pdf, jira)
Add it to docker/skills-manifest.txt:
1

pdf
Rebuild and deploy (the manifest is baked into the image):
1

make deploy REBUILD=1

Included Skills

Skill	Description
`yt`	YouTube transcript fetching and video search
`agent-browser`	Headless browser for JS-heavy/paywalled pages
`system-monitor`	CPU/RAM/GPU status check
`conventional-commits`	Format commit messages per convention
`github`	GitHub issues, PRs, CI runs via gh CLI
`summarize`	Summarize URLs, PDFs, images, audio, YouTube
`weather`	Weather lookup

Headless Browser

Mon, 01 Jan 0001 00:00:00 +0000

Headless Browser

OpenClaw includes a built-in browser powered by agent-browser running inside the gateway container. Chromium and its dependencies are installed at Docker build time via agent-browser install --with-deps.

No sidecar container is needed — the browser runs in-process within the OpenClaw gateway.

How It Works

The agent-browser npm package manages a local Chromium installation and exposes browser automation capabilities to OpenClaw skills. The gateway configures it automatically.

Verifying the Installation

1

make exec CMD="npx agent-browser --version"

The browser is ready when the gateway logs show agent-browser initialization complete.

Workspace Git Sync

Mon, 01 Jan 0001 00:00:00 +0000

Workspace Git Sync

Back up ~/.openclaw/workspace to a private GitHub repo automatically. Runs as a native cron job on the VPS, pushing to a configurable branch (default: auto).

Setup

Add to secrets/.env:

1
2
3
4


GIT_WORKSPACE_REPO=your-username/openclaw-workspace
GIT_WORKSPACE_BRANCH=auto
GIT_WORKSPACE_TOKEN=ghp_your_personal_access_token
GIT_WORKSPACE_SYNC_SCHEDULE=0 4 * * *

The cron job auto-enables when GIT_WORKSPACE_REPO is set. Deploy with make deploy.

Operations

1
2
3
4
5


# Trigger sync immediately
make workspace-sync

# Disable: remove or clear GIT_WORKSPACE_REPO from .env and redeploy
make deploy

Logs

The cron job writes to ~/openclaw/logs/workspace-sync.log on the VPS: