OpenClaw Asks for Your Inbox, Your Calendar, and $1 a Month

This week, a video from Tech in Asia, a regional technology publication, made the rounds online. ^[1] The video introduces a new product called OpenClaw, which the host describes as an “agentic autonomous AI.” The host identifies himself as an AI skeptic who found the product “intriguing.”

Here is what the video says OpenClaw does, in order. You can chat with it to organize your day, control your appliances, and — over time, the host says, citing one developer’s testimony — adapt to your daily routine as it gets to know you. ^[2] To do that, the software accesses your emails, your calendars, and “pretty much anything else” you choose to give it. ^[3] Let us pause on the phrase “pretty much anything else.” In my opinion, “pretty much anything” is doing a lot of work in a product description. In my opinion, the bar for “what we will read” being set by the user’s enthusiasm is not a bar at all. In my opinion.

The reassuring part, the host notes, is that the software runs locally on your device. ^[4] On paper, that means it does not phone home to a server farm somewhere. On paper, your data stays in your laptop. On paper, the house does not burn down. We are big fans of paper, here.

However — and this is the part where I put on my glasses — to actually run it that way, the user has to set up the entire infrastructure and backend themselves. ^[5] The host is candid about the implication. He says that if you are, in his words, “a bit of an idiot” during setup, you can expose yourself to security risks. Things like a hacker breaking into your messaging app, accessing your assistant, and instructing it to exfiltrate whatever data you have stored. ^[6] I appreciate the candor. I appreciate being told, in plain language, that the safe version of the product requires the user to not be the kind of person most products are designed for.

The host is upfront about the rest of the catch. For the system to be “completely bulletproof,” he says, you have to be a computer expert. ^[7] Then, in a moment I find genuinely charming, he says: “And that I am not.” ^[8] This is the part of a normal product launch where, in a press conference, an executive would say “we are confident in the product.” Instead, the presenter says: I cannot use it safely, and I am presenting it to you anyway. Tech in Asia has not yet responded to my request to clarify whether this is a sales strategy.

If being a computer expert is a non-starter, you can always pay. A cloud-based version is available, with prices starting from $1 a month. ^[9] One United States dollar. To give a third party read access to your inbox, your schedule, your appliances, and your routine. I did mathematics. That is a steal, in the sense that the data alone is worth considerably more.

Once you are set up, the host adds, your instance of OpenClaw can talk to other instances of itself online, on a social media website that the host describes as “ostensibly made for bots.” ^[10] Just like pretty much every other social media platform, then. Same energy. Same engagement metrics. Same trajectory.

The host also floats the possibility that one day, the AI might be able to “rent a human.” He does not elaborate. I will not speculate. I will simply note that a piece of software designed to organize your life, read your mail, and chat with its peers on a bot network is now being gently introduced to the idea of hiring a person. Make of that what you will.

The honest summary of OpenClaw, as presented, is this. The product offers to do helpful things with your digital life. To do them, it asks for broad access. To keep that access safe, you must already know how to keep broad access safe. If you do not, the company will hold your hand for the low, low price of $1 a month, after which it will be responsible for the rest. The training wheels cost a dollar. The bicycle costs your inbox.

I have no further comment. Neither, it seems, does anyone who has actually finished setting it up.