What is OpenClaw?

OpenClaw runs an AI agent as a persistent service and connects it to the places you already chat. Instead of opening a web app every time you want the agent, you message it in Telegram, Discord, or Slack, and it answers there. The agent keeps running on its host between messages, so it can hold context, remember past work, and act on a schedule.

The project is open source, which means the code is public and you run your own instance rather than renting access to someone else's. That appeals to developers who want an always-on assistant on hardware they control, with their own model provider behind it, and no third party sitting between them and the agent.

OpenClaw splits into a Gateway and nodes. The Gateway is the long-running process that holds the agent loop, its memory, its skills, and its outbound channels. It is the part that has to stay awake, which is why it lives on a host that does not sleep: a Linux box, a Mac, a Windows machine, a Raspberry Pi, a NAS, or a VPS.

Nodes are clients that connect to the Gateway. A phone, a browser, or a laptop pairs with the Gateway and acts as a control surface and, in some cases, a sensor that can offer the camera or microphone. The important consequence is that an iPhone alone does not run OpenClaw; it pairs with a Gateway running somewhere stable. That split is deliberate, because an agent needs a steady home so its memory and scheduled loops survive a phone going to sleep.

OpenClaw is a general agent host, so what you build depends on the skills and channels you give it. Common setups include a personal assistant that answers in Telegram, a coding helper that works on a repository, a monitoring agent that runs checks on a schedule, and a research agent that gathers and summarizes information on request.

Because the Gateway is always on, the most useful patterns are the ones that benefit from persistence. An agent that remembers your projects across days, runs a daily summary at 7am, or watches a system and pings you when something breaks all rely on a process that stays awake. That is the gap between a chatbot you open and an agent that is simply there.

OpenClaw is free and open source. There is no license fee for the software itself. What you pay for is the LLM provider behind it and the host it runs on, and both of those are your choice. You bring your own API key, so the running cost tracks your model usage rather than a subscription to OpenClaw.

That cost can be small. With a hosted provider doing the model work, the Gateway runs on a 1 GB VPS, and several budget hosts offer free tiers. You can also point OpenClaw at cheaper open-source model APIs to cut the per-token cost. The headline is that the framework adds nothing to the bill; the model and the host are the only line items.

OpenClaw, Hermes, and Claude Code sit at different layers, so they are not direct substitutes. OpenClaw is a gateway: its strength is routing an agent to channels and plugins. Hermes is a self-improving runtime: its strength is the agent itself learning skills from your work and keeping memory. Claude Code is a coding agent that runs in a terminal and is strongest at working directly in a codebase.

A practical way to choose: pick OpenClaw if your bottleneck is channels and plugin routing, pick Hermes if your bottleneck is the agent getting better at your tasks over time, and pick Claude Code if your bottleneck is hands-on code editing. They can coexist on one host, and Onepilot deploys OpenClaw and Hermes through a wizard while running Claude Code and Codex in its terminal, so the same app covers all of them.

You run OpenClaw from an iPhone in one of two ways. The first is to pair the iPhone as a node to a Gateway you already set up on a Mac or Linux box. The second is to use Onepilot, which provisions the Gateway for you on whatever host you can reach over SSH, then makes the iPhone the control surface.

With Onepilot the flow is a two-minute wizard: add the host, tap Deploy Agent and choose OpenClaw, pick one of 25 LLM providers and paste the key, pick a channel (Telegram, Discord, or Slack), and confirm. After that the iPhone gives you the host's real terminal, a syntax-highlighted file browser, git diffs, and cron, so you can both talk to the agent and operate the machine it lives on. The Gateway keeps running when the phone is locked.