OpenClaw

OpenClaw is an open-source AI assistant from the OpenClaw project. According to the official website, the assistant runs on your own machine and can be reached through WhatsApp, Telegram, or other chat apps. That makes OpenClaw more than an ordinary web chat: it is a self-operated connection layer between a personal gateway, selected communication channels, and the AI agents that an operator chooses to enable. The official website at openclaw.ai and the repository at github.com/openclaw/openclaw are the authoritative starting points for installation, versions, and security decisions.

Your gateway instead of a hosted account

The key difference is the operating model. OpenClaw is installed and managed on infrastructure controlled by the operator. The gateway receives messages from connected channels, associates them with an agent session, and can return replies or approved actions to the relevant channel. Actual data control therefore depends not only on the source code but also on the selected model connection, channel accounts, hosting environment, logs, and enabled tools. Before production use, determine which data stays local, which data is sent to a model provider, and how long messages, files, and logs are retained.

Chat channels and daily work

According to the provider, OpenClaw can connect several chat interfaces and channel plugins. Depending on the current documentation, these include Discord, Google Chat, iMessage, Matrix, Microsoft Teams, Signal, Slack, Telegram, and WhatsApp, among others. Channel access can make the assistant useful for personal notes, scheduling, research, summaries, and recurring tasks. Each connection still needs an individual permission review. A bot that may read messages does not automatically need permission to send, forward, delete, or modify account data. Channel and group permissions should be limited to what the task actually requires.

Coding and agent work

OpenClaw can provide access to AI agents that work with files, command-line tools, or development projects. This is useful when a controlled personal computer or server should be the working location for research, development, and automation. Every enabled capability also increases the impact of a compromised account, a manipulated message, or a malicious file in a project. Content from chats, websites, tickets, and repositories is data, not automatically valid instruction. Changes to files, dependencies, deployments, accounts, or external services should first appear as a reviewable plan and only run after explicit human approval.

Self-hosting and security boundaries

Self-hosted does not mean secure by default. Operators must protect gateway access, network exposure, the operating-system account, secrets, backups, logs, updates, and model connections. The official documentation describes a trusted single-operator setup as the starting point; teams and shared environments require deliberate hardening and clear roles. Important controls include separate accounts, minimal tool permissions, protected configuration files, regular updates, an inventory of enabled plugins, and an audit of the channels that are actually reachable. Emergency and financial actions should never be triggered solely by a chat message.

Who OpenClaw is for

OpenClaw is suited to technically capable individuals and teams that want to self-host a flexible personal assistant and deliberately control channels and agent capabilities. It is especially relevant for people who prefer local infrastructure, open-source software, and configurable automation over an exclusively hosted assistant. It is less suitable for anyone seeking a finished application with no ongoing administration, security review, or model and channel management. Before installation, review the official setup instructions, supported channels, and the project security guidance. OpenClaw is available as open-source software without a product purchase price; hosting, model, messaging-channel, and third-party costs remain subject to the respective providers' current terms.