The semantic CDN for AI agents. Crawl once, serve everyone.
"If 1,000 agents all need to read stripe.com/pricing today, they all crawl it separately. That's 1,000 redundant fetches."
GET /v1/som?url=stripe.com/pricing
One endpoint. Pass any URL. Get structured semantic data back.
Hot — Redis, sub-ms
Warm — SQLite, fast
Cold — Live crawl via Plasmate
Structured SOM JSON with compression stats and token savings. Ready for any LLM.
# Fetch the semantic representation of a page curl "https://cache.plasmate.app/v1/som?url=https://stripe.com/pricing" \ -H "Authorization: Bearer sk_som_..."
{
"url": "https://stripe.com/pricing",
"cached": true,
"cache_tier": "hot",
"som": {
"elements": 142,
"tree": [ /* semantic DOM nodes */ ]
},
"stats": {
"raw_html_bytes": 284000,
"som_bytes": 8200,
"compression_ratio": 34.6,
"tokens_saved": 68000
},
"crawled_at": "2025-03-23T18:42:00Z"
}
| Site | Compression |
|---|---|
| google.com | 1,400× |
| duckduckgo.com | 33× |
| wikipedia.org | 24× |
| stripe.com/pricing | 34× |
| docs.python.org | 18× |