Guides

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: ...

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 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 The stack includes a chromium container (chromedp/headless-shell) so OpenClaw can use the browser tool on the VPS — useful for scraping JS-heavy or paywalled pages via the agent-browser skill. How It Works Chrome’s CDP endpoint rejects WebSocket connections where the Host header is not an IP address or localhost (DNS rebinding protection). To work around this without a proxy, the chromium service is assigned a static IP (172.20.0.10) on a custom Docker bridge subnet (172.20.0.0/24). OpenClaw connects directly to http://172.20.0.10:9222 — Chrome accepts IP addresses unconditionally. ...

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: ...