Everything OpenClaw can do — documented clearly

Skills are the core of what OpenClaw does. Each skill adds a discrete capability, and ClawHub is the distribution layer where teams discover, install, and update them.

• Install skills with a single command from ClawHub.
• Write custom skills in JavaScript or TypeScript.
• Chain skills together for multi-step automations.
• Version-lock skills to avoid breaking changes.

SOUL.md shapes the agent's persona and defaults. MEMORY.md stores durable context. Together they define how the assistant behaves over time and across sessions.

• SOUL.md controls tone, persona, and default behaviors.
• MEMORY.md persists facts across sessions.
• Both files support Markdown formatting and code blocks.
• Changes take effect on the next full agent restart.

OpenClaw connects to major messaging platforms through adapters. Each integration has its own authentication model, operational limits, and debugging path.

• WhatsApp via Baileys with no API key required.
• Telegram, Discord, Slack, and Signal support.
• Custom webhook adapters for unsupported platforms.
• Per-platform auth and rate-limit considerations.

The main config file controls providers, rate limits, ports, and logging behavior. This overview highlights the categories that usually need deliberate production choices.

• AI provider selection for hosted and local models.
• Rate limiting and concurrent session controls.
• Logging levels and output destinations.
• Environment variable mapping and secret handling.

OpenClaw can support multiple agents with separate instructions, memories, and skill sets while sharing the underlying host. The orchestration pattern matters for isolation and operability.

• Separate persona and memory per agent.
• Per-agent skill sets and overrides.
• Shared infrastructure with explicit isolation boundaries.
• Operational planning for ports, logs, and ownership.

You can run OpenClaw against Ollama, LM Studio, or any OpenAI-compatible endpoint. Local models trade some convenience for privacy, cost control, and deployment flexibility.

• Ollama and LM Studio support out of the box.
• Any OpenAI-compatible endpoint can work.
• Model fallback chains improve reliability.
• Per-skill model routing helps control cost.