Harness profile
OpenClaw: an open-source personal assistant that lives in your chat apps
OpenClaw is an open-source, self-hosted AI assistant that runs on your own machine and answers you in the chat apps you already use, such as WhatsApp, Slack, Teams and Telegram. It is best for one person or a small trusted team who want an always-on assistant with scheduled jobs, skills and tools under their own control.
Explained from public sources. Mai has not built this yet.
On this page
In plain English
OpenClaw is an AI assistant you install on your own computer or server rather than use as a website. Once it is running, you talk to it through the messaging apps you already have, so a WhatsApp message or a Slack mention reaches it the same way a text to a colleague would. It can run jobs on a timer, use tools such as a browser or a terminal, and learn reusable routines. Your settings, memory and credentials stay on the hardware you run it on.
How it works
The centre of OpenClaw is the Gateway, a single long-running process on your machine or a server. The docs describe it as the source of truth for sessions, routing and channel connections. Chat apps connect to it through channel plugins, and the command line, a terminal interface, a web Control UI and companion apps for macOS, iOS and Android all connect to it as clients.
It is open source under the MIT licence and runs on Node.js (the docs recommend Node 26, with Node 24 still supported). You install it with a one-line script or from npm, then run openclaw onboard to set up the Gateway as a background service. Configuration lives in a JSON file under ~/.openclaw/. It works with hosted and local model providers, and the README describes models and agent harnesses as swappable plugins; you bring an API key for the provider you choose.
Channels: WhatsApp, Slack, Teams, Telegram. Then Gateway. Then Agents: Main agent, Specialist agents. Then Reach: Built-in tools, Skills, MCP servers. Then Exec approval.
Several pieces sit on top of the Gateway:
- Agents and routing. One Gateway can host several agents, each with its own workspace, state folder and session history. Bindings decide which agent answers which WhatsApp number or Slack workspace, and the most specific match wins. Agents can delegate bounded work to sub-agents if you allow it.
- Tools. A large built-in set covers shell commands, file edits, web search, a browser, messaging, media generation and scheduling. MCP servers can be added over stdio, Streamable HTTP or SSE, and their tools sit under the same tool policy.
- Skills. A skill is a folder with a
SKILL.mdfile that tells the agent when and how to use its tools. Skills load from the workspace, shared folders, bundled sets and ClawHub, the public registry. A feature called Skill Workshop lets an agent save and update its own skills, keeping the previous version so a change can be undone. - Automation. A built-in scheduler runs one-off reminders, intervals and cron expressions, and can also start on an incoming webhook. What used to be a separate heartbeat is now an ordinary scheduled job. Standing orders in
AGENTS.mdgive the agent lasting instructions that are added to every session.
What it does best
OpenClaw is strongest as a personal or small-team assistant that is always on and always reachable. The breadth of channels is the standout: the same agent can answer in WhatsApp, Signal, Teams and a dozen others without separate builds. Scheduling, webhooks and standing orders make it proactive rather than only reactive, and the skills system, including skills the agent writes for itself, gives it a way to get better at repeated tasks.
Keeping everything on your own hardware is a real advantage for people who do not want their messages, memory and keys held by a vendor. On telemetry, the README says the default is a daily version check, with anonymous feature statistics only if you opt in.
Where it stops
The security guide is clear that one Gateway is one trust boundary. It is designed for a single operator or a team that trusts each other, not as a shared service for people who should not see each other’s data. Mixed-trust use needs separate Gateways and credentials.
It is an assistant platform, not a library. If you want to embed an agent loop inside your own application and control every step in code, a software development kit is a closer fit. It also assumes someone is comfortable running and updating a Node.js service, managing a config file and keeping an eye on a growing set of channels and plugins.
Where it goes wrong
The docs flag the risks plainly. Tools run on the host for the main session unless you configure sandboxing, and an agent’s workspace is a default folder rather than a hard wall. Exec approvals default to allowing commands on the Gateway host, so a careful setup has to tighten them. Agents with messaging access can, by default, send across conversations and providers until you restrict that. Third-party skills are to be treated as untrusted code and read before use. The guide recommends running openclaw security audit regularly to catch settings drifting away from a safe baseline.
How Mai AI uses it
I would point someone to OpenClaw when they want a capable personal assistant for themselves and are happy to run it on their own machine. For a business system serving several staff and customers, its own guidance about one trust boundary per Gateway is the deciding factor, and I would build the channel layer to separate those boundaries from the start.
The ideas I take from it are the Gateway as a single control point for every channel, and pairing unknown senders before they can talk to the agent at all. Both carry straight into agents in the channels people already use. Its layered exec approvals are a good reference for human approval gates, and its frank security guide is worth reading alongside agent safety. For how these pieces fit into a system built around one business, see how Mai AI designs custom AI operations systems.
Questions and answers
Is OpenClaw free?
Yes. OpenClaw is open source under the MIT licence and is stewarded by the OpenClaw Foundation, which says there is no paid tier or hosted service. You still pay for the model provider you connect it to.
What chat apps does OpenClaw work with?
The docs list built-in support for Discord, Google Chat, iMessage, Matrix, Microsoft Teams, Signal, Slack, Telegram, WhatsApp, Zalo and a web chat, with more channels available as plugins.
Does OpenClaw support MCP?
Yes. You can add MCP servers over stdio, Streamable HTTP or SSE, and their tools become available to your agents under the same tool policy as built-in tools. A separate command can also expose OpenClaw conversations to another MCP client.
Is OpenClaw safe to use with several people?
The security guide treats one gateway as one trust boundary: a single operator or a team that trusts each other. For people who should not share access, it recommends separate gateways and credentials, ideally on separate users or hosts.
Does OpenClaw run commands without asking?
It depends on configuration. Exec approvals let you deny host commands, allow only a list, or ask on anything outside the list, and when no approval screen is reachable the default fallback is to deny. Tools run on the host for the main session unless you turn on sandboxing.
Patterns it is known for
- PatternAgents in your channels
You put an AI agent in email, WhatsApp, Slack or Teams by connecting it to each platform's official messaging interface, so people message it where they already work. One core agent sits behind thin adapters that handle each channel's rules, such as WhatsApp's 24-hour reply window.
- PatternHeartbeats and schedules
AI agents work without being asked when something other than a person wakes them: a schedule, a regular heartbeat check or an incoming event. Each wake-up gives the agent a fixed job, a time budget and a way to report back, so it can chase, prepare and flag things before anyone thinks to ask.
- PatternSelf-improving skills
Self-improving agents get better by writing down what worked as reusable skills: short procedures, checklists or scripts they load next time a similar task comes up. The model does not change; the library of procedures around it does, ideally with a person reviewing each change.
- PatternRouting and specialist agents
Routing means one agent reads each request and hands it to the specialist best placed to handle it, such as finance, operations or sales. Each specialist has its own instructions, tools and permissions, which keeps answers sharper and access tighter than one agent trying to do everything.
- PatternThe tool layer and MCP
AI agents reach real systems through tools: named functions such as "find invoice" that the model asks to run, while ordinary code does the actual work. MCP, the Model Context Protocol, is an open standard for offering those tools and data to any AI application in the same way.
- PatternHuman approval gates
Human approval gates sort every action an AI agent can take into three levels: draft only, needs a person to approve, or allowed to run automatically. The agent prepares, a person signs off anything that matters, and the system refuses any action that does not match what was approved.
- PatternAgent safety
You make an AI agent safe by limiting what it can do, not by trusting what it says. Give it the narrowest tools and permissions, run risky work in a sandbox, require a person to approve consequential actions, and treat every email, document and web page it reads as possibly hostile.
Related entries
- Harness profileHermes Agent
Hermes Agent is an open-source AI agent from Nous Research that runs on a server or your own machine, talks to you through the terminal or chat apps, and writes reusable skills from the work it does. It suits people who want a long-lived agent that remembers, runs scheduled jobs and improves with use, on any model provider.
Sources
- openclaw on GitHub (README)github.com
- OpenClaw documentation homedocs.openclaw.ai
- Securitydocs.openclaw.ai
- Automationdocs.openclaw.ai
- Multi-agent routingdocs.openclaw.ai
- Toolsdocs.openclaw.ai
- Skillsdocs.openclaw.ai
- MCPdocs.openclaw.ai
- Exec approvalsdocs.openclaw.ai