OpenClaw

OpenClaw is a free, open-source AI agent from Peter Steinberger and the Vienna-based OpenClaw team that runs 24/7 on your own machine and talks back through the messaging apps you already use. Install it via CLI, point it at any LLM, and it becomes a persistent assistant that remembers you across every session, runs scheduled tasks, drives your browser, and writes its own new skills on demand.

What OpenClaw does:

• •Persistent memory stores your context locally, recalls it on every conversation, and builds a long-running model of your preferences across days and weeks.
• •9 messaging integrations chat with the agent on WhatsApp, Telegram, Discord, Slack, Signal, iMessage, Teams, Google Chat, and Matrix without installing any new app on your phone.
• •Heartbeat check-ins the agent proactively pings you on a schedule with briefings, reminders, and status updates rather than waiting for prompts.
• •Browser control navigates real websites, fills forms, and extracts data autonomously, so research and form-filling tasks run without manual handoff.
• •System and shell access reads and writes files, runs shell commands, and executes scripts on your machine, which is the foundation for the devops and automation use cases.
• •Self-writing skills describe a capability you want, OpenClaw writes the code, installs it, and runs it, then offers to share it back to ClawHub.
• •ClawHub registry community skill marketplace where every published skill is scanned by VirusTotal before listing, so you can install community work without auditing each one yourself.
• •MCP support via mcporter a lightweight bridge exposes any Model Context Protocol server to OpenClaw, so MCP tool servers (databases, file stores, third-party APIs) plug in without custom code.
• •CLI configuration the entire agent is configured from the command line and a YAML file, which is the open-source, hackable counterpart to a closed cloud agent.
• •Open-source MIT licence the full agent is open-source under the MIT licence on GitHub, so you can fork, audit, and self-host with no commercial restrictions.

Pricing:

• •Open-source self-hosted $0: free under MIT licence, you only pay for your own VPS ($6 to $15 per month) plus LLM API tokens.
• •Light usage total $6 to $13/month: small VPS plus light LLM token spend on a single connected provider.
• •Heavy usage total $100 to $200+/month: heavier LLM model use, longer browser sessions, and bigger memory contexts on the agent's own provider bill.

Limitations:

• •Self-hosted setup is not plug-and-play requires Node.js 22+, WSL2 on Windows, and manual model configuration, with reviewers reporting a 45-minute minimum even for technical users.
• •API token costs can spike without warning one developer reported a $623 monthly bill from unmonitored agent activity, so spending caps and provider-side usage alerts are essential.
• •Windows needs WSL2 first Windows users must install the Windows Subsystem for Linux as a prerequisite, which is an unexpected step for most small business owners on Windows machines.
• •No genuine strategic judgement open-ended tasks like planning a marketing campaign produce shallow results, so OpenClaw follows instructions well but does not replace real human strategy.
• •Async chat is not real-time interactive because the agent runs on heartbeats and message events, it is not a substitute for a live coding pair like Cursor or Claude Code in interactive flow.