OpenClaw for Real Estate Agents: Setup, Use Cases, and What Can Go Wrong

What OpenClaw is

OpenClaw is an MIT-licensed AI assistant that runs on your own computer and answers you in the chat apps you already use.

A nonprofit, the OpenClaw Foundation, runs it. The project says it has no paid tier, no hosted service and no token.

Its creator, Peter Steinberger, joined OpenAI in February and said OpenClaw would stay open and independent. OpenAI is now a donor to the foundation, not an owner.

OpenClaw 2.0 came out Aug 30, 2026. It made setup easier by reusing ChatGPT or Claude sign-ins you already have, and it added shared sessions so another person can join the agent's work.

What it costs

The software is free; you pay for the AI model behind it and, if you want it running all the time, the computer it runs on.

• The model: an existing ChatGPT or Claude sign-in, an API key billed per use, or free local models through Ollama.
• The computer: a Mac or Linux machine, or Windows through WSL2. A Mac mini is a popular always-on choice, but a small server works too. You only need a Mac for iMessage.
• Always-on: a laptop has to stay awake. The docs suggest a small server for around-the-clock use.
• Hosting: some companies sell hosted OpenClaw for a monthly fee. Those are third parties, not the OpenClaw project.
• For shared business use of Claude, OpenClaw's own docs recommend an Anthropic API key rather than a personal sign-in.

What an agent can use it for

The best real estate example so far is Daniel Foch, Chief Real Estate Officer at Valery.ca, whom Inman profiled in March.

Inman reported that his assistant handles roughly 70 to 80 percent of the admin work that used to take 20 to 25 hours of his week. It connects to Follow Up Boss and kvCORE through their APIs.

He can text it to move a contact into a nurture sequence. He is now turning his setup into a product for agents called Homie.

Inman lists jobs like these among the ones his assistant handles:

• Update contacts and create follow-up tasks in your CRM.
• Schedule showings and manage your calendar.
• Organize documents for each transaction.
• Draft listing presentations, home evaluations and newsletters.

What can go wrong

OpenClaw's own docs say prompt injection is not a solved problem and sandboxing is off by default.

Prompt injection means a booby-trapped email, web page or document can carry instructions the agent follows, even if you're the only one who messages it. For an agent whose inbox is full of strangers' emails, that is the risk to take seriously.

• Plugins and skills run as trusted code. In February, researchers found 341 malicious skills on ClawHub, many installing a password stealer.
• Early versions left many installs open to the internet: over 30,000 by Bitsight's count and over 135,000 by SecurityScorecard's, as reported by The Register, both in February. Regular installs now listen only on the same machine.
• The Register's review of 2.0 found stored secrets are not encrypted at rest.
• By default, an agent that can send messages can send across conversations and apps. For you, that could mean a client text you never wrote.
• The 2.0 upgrade broke some setups, with lost automations and sign-in problems (InfoQ). Back up first.
• Cisco researchers, quoted by Inman: "From a security perspective, it's an absolute nightmare."