Limits & Costs
The plugin is free and there is no paid tier. The running costs land on your Cloudflare account, and for most sites they land inside the free allocations. This page is about where the ceilings actually are.
What you pay us
Nothing. There is no licence, no metering, and no per-message fee, because we are not in the message path and have nothing to meter.
What you pay Cloudflare
Three services are involved, and all three have free allocations: Workers (the endpoint), Workers AI (the models), and Vectorize (the index). A typical small-to-medium site (a few hundred pages, a few hundred conversations a month) fits inside them.
Cloudflare publishes and occasionally revises the exact figures, so check their pricing pages for current numbers rather than trusting a figure written here. The useful thing to understand is which dimension you will hit first.
The dimension that usually binds: index size
Vectorize's free allocation is measured in stored vector dimensions. At the 768 dimensions this plugin uses, that works out to a few thousand chunks. Since a chunk is roughly 1,500 characters, a site with a few thousand pages' worth of indexed prose is in the region where it matters.
There is no warning yet
This release does not warn you as you approach the Vectorize ceiling. If you run a large site, keep an eye on the Vectorize section of the Cloudflare dashboard after the first index completes. A friendlier in-plugin warning is a known gap.
If you are close to the limit, the first thing to do is not to upgrade but to index less. Most sites have a large tail of content that adds nothing to a support chatbot: tag archives, old announcements, thin pages. Narrowing the selected post types and excluding individual items is usually enough, and it improves answer quality at the same time.
Message volume
Every message costs one embedding call and one answering call. The answering model is much the larger of the two, so a bigger model burns the allocation faster. If volume is your constraint, a smaller answering model is the lever.
Requests
The Worker is hit once per message, not once per page view, so request volume is rarely the binding constraint.
Built-in ceilings
These are enforced by the plugin regardless of your Cloudflare plan:
| Limit | Value |
|---|---|
| Messages per IP per minute | 20 |
| Conversation history sent to the model | 20 messages |
| Maximum message length | 4,000 characters |
| Content chunk size | ~1,500 characters |
| Passages retrieved per question | 3 |
| Suggested questions in the widget | 6 |
The rate limit is not configurable in this release. Making it so is on the roadmap.
If you outgrow the free plan
Cloudflare's paid Workers plan raises all of these ceilings, and you buy it from Cloudflare directly. Nothing about the plugin changes and there is nothing to upgrade on our side: the plugin works identically on either plan.
Keeping costs down
- Index the content that answers questions, not everything you have ever published.
- Exclude thin and outdated pages. This cuts index size and improves answers at once.
- Use suggested questions to steer visitors toward what the bot answers well, which reduces the back-and-forth of failed attempts.
- Give the bot a human support link so a visitor it cannot help leaves rather than trying twelve more phrasings.