When I started with OpenClaw, I spent days searching for “the complete list of everything this platform can do.” I found blog posts, GitHub repos, Discord threads, YouTube tutorials — scattered across dozens of sources, most of them incomplete or outdated.
So I made the map myself. Everything I’ve found in eight months of daily use, organized by what you’ll actually need it for.
This isn’t a marketing overview. It’s a practitioner’s field guide — what exists, what’s good, what’s mediocre, and what you can skip.
Core Platform
OpenClaw itself is the orchestration layer. It handles sessions, tool management, model routing, scheduling (cron), and the integration pipeline. Think of it as the central nervous system that connects everything else.
It runs on anything from a Raspberry Pi to a cloud VM. I run mine on a $20/month VPS with 4GB RAM and it handles everything I throw at it. Heavier workloads (high concurrency, large context windows) benefit from more RAM.
Configuration is YAML-based. This is simultaneously OpenClaw’s greatest strength (infinitely customizable) and biggest friction point (you need to understand the config format). The documentation covers most settings, and the community Discord fills the gaps.
AI Models: What Works With OpenClaw
OpenClaw is model-agnostic — it works with any LLM through API connections:
Anthropic Claude is what I use for most tasks. Strong reasoning, good at following complex instructions, handles long documents well. Claude is my recommendation for anyone who wants one model for everything.
OpenAI GPT-4o is the alternative. Similar capability to Claude, sometimes better at creative tasks, sometimes worse at structured analysis. Honestly, for most real-world tasks, the difference doesn’t matter enough to agonize over.
Local models via Ollama. Run Llama, Mistral, or other open-source models on your own hardware. Great for privacy-sensitive workloads and for keeping costs at zero. Quality is a step below the top API models, but the gap is shrinking fast.
My recommendation: Start with one API model (Claude or GPT-4o). Add a local model later for simple tasks and cost optimization. Don’t overthink the model choice — they’re all good enough.
Skills: The Plugin Ecosystem
Skills are OpenClaw’s plugin system. Each skill adds a specific capability. The essential ones:
Tier 1 (install immediately):
– Web search — gives your agent internet access
– GitHub — full repo/issue/PR management
– File system — read/write files
– Browser automation — control web browsers
– Summarization — summarize URLs, docs, videos
Tier 2 (install when needed):
– Coding agent — delegate complex coding tasks
– Database query — natural language database access
– PDF tools — read and extract from PDFs
– Weather — add weather data to briefings
– Health check — monitor server/site status
Tier 3 (niche but useful):
– Video frames — extract frames from video
– Tmux — control terminal sessions remotely
– Peekaboo — capture and automate macOS UI
The ClawHub marketplace has community-contributed skills. Quality varies — some are excellent, some are abandoned experiments. Check the last update date and star count before installing.
Integrations: Connecting to Your World
Messaging platforms: Discord, Slack, Telegram, WhatsApp, iMessage, Signal, Line, Feishu, Google Chat. Discord and Telegram are the most mature integrations. WhatsApp works but has limitations due to Meta’s API restrictions.
Development tools: GitHub (native), GitLab (via API), various CI/CD platforms. The GitHub integration is deep — issues, PRs, reviews, actions, and code browsing.
Productivity tools: Notion, Google Workspace (via third-party CLIs), calendar systems. Integration depth varies — some are read/write, some are read-only.
Infrastructure: SSH for remote server management, Docker for containerized deployments, various monitoring tools.
Community Resources
Discord is the most active community. Real-time help, workflow sharing, and skill announcements. The #help channel is responsive — most questions get answered within an hour during active hours.
GitHub has the source code, issues tracker, and discussions. This is where you report bugs, request features, and contribute code.
Documentation site covers the technical details: configuration reference, API documentation, and setup guides. It’s thorough but sometimes lags behind the latest features.
Blog posts and tutorials (like this one) provide real-world perspectives and practical advice that documentation doesn’t cover. Search for specific use cases rather than general overviews.
The Learning Path I Recommend
Week 1: Install OpenClaw, connect one messaging platform (Slack or Discord), and one AI model. Send it messages. Get comfortable with the basics.
Week 2: Install 3-5 essential skills (web search, file system, GitHub). Build your first simple automation — a morning briefing or a Slack bot that answers questions.
Week 3: Set up cron jobs for scheduled tasks. Start with one daily automation.
Week 4: Add monitoring (even basic logging is better than nothing). Connect additional integrations as needed.
Month 2+: Explore advanced features — multi-agent workflows, custom skills, complex integrations. By this point, you’ll know what you need because you’ll be hitting specific limitations.
The ecosystem is large and growing. You don’t need to learn all of it — you need to learn the parts that solve your specific problems. Start narrow, expand as needed, and don’t install skills “just in case.” Every unused skill is clutter you’ll have to maintain.
🕒 Last updated: · Originally published: January 1, 2026