BYOBot is an AI system. This page explains, in plain language, where AI sits in everything we make: the product you type into, the articles you read on this site, and the images on them. Nothing here is hidden in a footnote. If you only read one line, read this one: you are interacting with software, not a person, and the writing on this site is machine-written.
This page also carries the disclosures we owe you under Article 50 of the EU AI Act (Regulation (EU) 2024/1689), the UK and EU consumer transparency rules, and the plain honesty we'd want if we were the ones reading.
The chat on the BYOBot homepage is not staffed. Every reply you get is generated by a large language model in real time. There's no human reading your messages as you type them, no support agent behind the cursor, and no queue you're waiting in. When you press send, your text goes to an AI model and the answer comes back.
We label this in the product itself, not only here. You'll see an AI notice in the interface before you type your first message.
You describe a repetitive task in your own words. BYOBot asks follow-up questions, then writes an automation spec: a portable, plain-markdown document that describes the task step by step so an AI tool can carry it out. You take that spec and paste it wherever you want it to run.
Here's the part people miss. BYOBot does not run your workflow. It doesn't connect to your email, open your CRM, touch your files, or execute anything on your behalf. It writes the instructions. You decide where they go and whether they run at all. That gap is deliberate, and it's the main reason BYOBot sits outside the higher-risk categories of the AI Act: the system produces a document for a human to review, not an action taken in the world.
Human in the loop, by design. Every spec BYOBot writes lands in front of you before anything happens with it. Read it. Change it. Throw it out. The tool has no way to act without you moving its output somewhere.
| Where | Provider | What it does |
|---|---|---|
| Chat and spec building | Anthropic (Claude) | Reads your description, asks questions, writes the automation spec |
| Web lookups inside chat | Anthropic web search | Pulls current information when your question needs it |
| Articles on /ai-news | Anthropic (Claude) | Researches and writes the published copy |
| Article images | Generative image models | Produces the illustration at the top of each article |
Model versions change as providers ship new ones. We don't repost this page for every version bump, but the providers listed above are the full set. If that ever changes, this table changes with it.
Everything published under /ai-news is generated by an automated publishing system. Not AI-assisted. Not lightly edited. Written by a machine, on a schedule, and posted without a person reading that specific piece first.
Luke Grace sets the rules the system writes to: the topics it's allowed to cover, the sources it draws from, the tone, and the hard limits on subject matter. What the system does inside those rules on any given Tuesday is its own. We think that's a reasonable way to run a news desk at our size, and we think you deserve to know it's how the sausage is made rather than finding out later.
What that means for you as a reader:
Each article carries this disclosure in three places: a notice above the author card, a marker in the byline, and machine-readable metadata in the page source so crawlers, platforms, and other AI systems can detect it without a human in the way.
Article 50(2) of the AI Act asks that synthetic content be marked in a machine-readable format. Every AI-generated article on this site carries:
<meta name="ai-generated" content="true"> tag in the page headgenerator meta tag naming the automated systemdigitalSourceType value trainedAlgorithmicMediaThese are the conventions that exist today. When better standards land, we'll adopt them.
Your messages go to Anthropic's API to generate a response. We don't store your conversation on our servers, we don't train any model on it, and we don't sell it. We have no account system, so there's no name or email attached to it in the first place. The only thing we keep is a random ID your browser generates, used to count your build credits.
Anthropic retains messages under their own privacy policy. The full picture of what we collect is on our privacy policy.
One practical note: don't paste secrets. API keys, passwords, client records, medical details, anything covered by an NDA. Describe the shape of your workflow instead of pasting the sensitive contents of it. A good spec doesn't need your real data to be useful.
Being useful and being honest about limits are the same job. BYOBot is weak in predictable places:
We assess BYOBot as a limited-risk AI system under the Act's risk tiers, which triggers transparency obligations rather than the conformity assessments required of high-risk systems. Our reasoning, stated openly so you can disagree with it:
We're a small team and this assessment is ours, not a regulator's finding. If you believe we've classified it wrong, we want the email.
When you take a spec and run it somewhere, you become the deployer of that automation. A few things follow from that, and none of them are meant to scare you off:
You can ask us what we hold about you, ask us to delete it, and object to how we use it. Since we hold almost nothing and none of it is linked to your identity, the answers are usually short. Email [email protected] and we'll respond within 30 days. If you're in the EU or UK and you're not satisfied with our answer, you can complain to your national data protection authority.
The AI Act phases in over several years and the guidance keeps moving. We'll update this page as our obligations and our own systems change, and the date at the top will tell you when we last did.
Questions about how we use AI, a correction to an article, or a disagreement with anything above: [email protected].