OpenClaw Platform Overview: Features & Capabilities

OpenClaw is easiest to understand as a Gateway that connects conversations to model providers and tools. The Gateway owns sessions, routing, approvals, and channel connections. A local Gateway does not automatically mean that inference is local; the provider configuration decides where a request is processed.

The pieces that matter

• Gateway: the service that receives requests, stores session state, and coordinates work.
• Channels: the messaging surfaces that can send and receive replies.
• Providers: hosted or local model endpoints configured separately from channels.
• Skills and plugins: extra instructions or code that expand what a session can do.
• Placements: the local host, paired device, or prepared worker where a task runs.

Follow one request

A message enters through a channel, is assigned to a session, reaches the selected provider, and may call an approved tool before the reply returns to the source conversation. Each step has a different failure mode. Start with Gateway status, then inspect the provider and channel separately.

openclaw --version
openclaw gateway status
openclaw doctor
openclaw security audit

Choose a boundary before a feature

Keep a personal Gateway and a shared team Gateway separate when the users do not share a trust boundary. Pair devices deliberately, gate group messages, and review skills or plugins before granting filesystem, network, browser, or process access.

Use the getting-started guide for installation, the provider reference for inference choices, and the security guide for exposure and access control. Those pages change with the release train; record the version you checked.