What an AI API key adds to Ariadne’s matching, and what happens if you don’t have one.
What it adds
Without an AI key, Ariadne still matches posts well: shared tags, shared categories, title overlap, and local text similarity all work normally, with no external service involved. An AI key adds one more layer on top, understanding what a post is actually about, not just which words it contains.
Two posts can be strongly related in meaning while sharing almost no vocabulary. A piece about “reducing server response time” and one about “why your website feels slow” are the same topic in different words. Local text matching alone won’t connect them reliably; semantic matching will.
Do you need one?
If your content is well-tagged and categorized, and posts on the same topic tend to use similar language, local matching alone (the Free tier) already performs well. An AI key is worth adding when:
- Your site covers a topic from many different angles or in different phrasing
- Tags and categories aren’t consistently applied across your archive
- You want the related-list quality to keep improving as your matching weights shift, without depending entirely on exact wording
How it works
Ariadne calls Voyage AI exactly once per post: when that post is saved, and only if its content has actually changed since the last time. Nothing is ever called while a visitor is viewing your site, so the cost is tied to how much you publish and edit, not how much traffic your site gets.
What it actually costs
Ariadne uses Voyage AI’s voyage-4 model by default, priced by Voyage at $0.06 per million tokens as of this writing. Every new Voyage account also gets 200 million tokens free, which is enough to cover the entire initial embedding backfill for almost any WordPress site, however large.
Each post sends at most 8,000 characters (roughly 2,000 tokens) to Voyage, capped automatically no matter how long the post actually is, so the cost per post has a hard ceiling even on very long articles.
Rough estimates by site size
These assume every post hits the 2,000-token cap, so they’re a worst case, not a typical one:
- Small site (200 posts, ~10 published or edited a month): the one-time backfill is about 400,000 tokens, well inside the free allowance. Ongoing cost after that: a fraction of a cent a month.
- Established site (5,000 posts, ~50 a month): backfill is about 10 million tokens, still fully covered by the free 200 million. Ongoing cost: well under $0.01/month.
- Large publisher (50,000 posts, ~500 a month): backfill is about 100 million tokens, still within the free allowance. Ongoing cost after that: roughly $0.06/month.
- Very large site (200,000+ posts): the one-time backfill can exceed the free 200 million tokens. A 400-million-token backfill costs about $12 total, one time, at the overage rate.
To put a hard ceiling on spending, set Settings → AI & API → Daily request limit. Once that many requests have been made in a day, AI summaries wait until the next day. One request covers a batch of posts, so even a small limit goes a long way. 0 means no limit.
This is billed directly by Voyage AI to your own account and card. Ariadne and Uwawp never see or handle that payment. See privacy and external services for how the connection works, or Voyage AI’s own pricing page for current rates, since pricing can change.
Setting it up
Once you’re on a Pro or Pro+ plan:
- Get an API key from Voyage AI.
- Add it either as
define( 'ARIADNE_VOYAGE_API_KEY', 'your-key-here' );inwp-config.php(recommended, keeps it out of the database), or paste it into the field on Settings → Ariadne Related Posts. - Save. Existing posts pick up semantic matching gradually, in the background, without you needing to trigger anything. A fresh install processes its whole archive the same way, a few posts at a time.
If you’d rather speed that initial pass along, the Post Status screen has a manual “catch up now” action. See command-line automation for a scriptable equivalent.