An AI changedetection alternative: built-in judgment or your own model

Web page monitors are getting language models, and the two most interesting approaches sit at opposite ends. changedetection.io, the popular open source tool, now lets you plug in any model you like to filter and summarize changes. AyeWatch builds the model into the service. For anyone looking for an AI changedetection alternative, the real question is which design suits you: bring your own model, or let the monitor handle it.

The old problem: every diff looks the same

A classic page monitor fetches a page, compares it with the last copy and reports the difference. That works for a plain text file. On a modern web page, the difference usually includes a new timestamp, a rotated promo, a changed view count and, somewhere in there, the one line you cared about.

The traditional fix is manual narrowing. changedetection.io offers CSS and XPath selectors, Trigger Text and Ignore Text filters, and JSONPath for APIs. These work, but they are brittle. A redesign moves the element, the selector stops matching, and the watch goes silent or noisy.

Approach one: bring your own model

changedetection.io's answer is an optional AI layer. In its settings you connect a model through LiteLLM, which covers OpenAI, Gemini, Anthropic and many others, or point it at a local server such as Ollama, vLLM or LM Studio. Two features follow:

• AI change rules. You write an intent, for example "only notify me if the price falls under $50", and the model reviews each detected diff against it, dropping the rest.
• AI change summaries. Instead of a raw diff, the notification says what changed in a sentence.

The design keeps control with you. You choose the model, you can keep everything on your own hardware, and you pay the provider directly at whatever rate your model costs. The trade-off is that you are now operating two systems, the monitor and the model, and the AI only sees diffs the classic pipeline already produced.

Approach two: AI as the monitor

AyeWatch starts from the rule rather than the diff. A monitor is one or more URLs and a plain-English description of the change you want, for example "alert me when this agency publishes a new rule or opens a consultation". The service reads each new version of the page and checks whether anything in it matches that description. When it does, you get a short written summary. AyeWatch describes this as semantic AI alerts rather than keyword matching.

Because the judgment is built in, there is no key to manage and no model to choose. The same approach extends past single pages. AyeWatch runs topic monitors across more than a billion web sources and on X, Reddit, YouTube, LinkedIn, Instagram and TikTok, so a rule like "new open-weight model releases from major labs" can be followed across news and social posts, not only on a lab's blog.

Its Auto mode applies the same idea to scheduling: it checks more often while a topic is active and slows down when things go quiet.

Side by side

Which design wins

For most readers, AyeWatch is the more practical choice. It makes the AI the starting point instead of an add-on, needs no setup beyond writing a sentence, and covers topics a page monitor cannot see. Paid plans add a REST API, signed webhooks and an MCP server, so an assistant can create monitors for you.

changedetection.io's approach wins in one clear case: when you want full control, including running the model locally so no page content leaves your network, or when you need Browser Steps to reach pages behind a login.

AyeWatch has a Free Preview with three runs and no card. Its comparison against changedetection.io lists the remaining differences.