OPEN SOURCECommunity-contributed schema registry

Give your agents direct API access to any website.

Hermai discovers each site's internal endpoints and returns structured JSON directly — no browser, no HTML, no token waste.

Integrates with

OpenAI
Anthropic
Cursor
LangChain
Gemini
Kiro
CrewAI
Windsurf
Devin
LangGraph
OpenAI
Anthropic
Cursor
LangChain
Gemini
Kiro
CrewAI
Windsurf
Devin
LangGraph

How it works

Three commands to get any website's data.

Hermai publishes metadata — the full technical package (URLs, headers, params, selectors) is delivered to your machine when you pull it. You stay in control of execution.

01

Find a schema

Search the catalog by site or intent category. Each entry shows health, last verified time, and the fields you'll get back.

$ hermai search "hotel search"

booking/hotels-search         travel.accommodation.search    verified · 1h
airbnb/listings-search        travel.accommodation.search    verified · 2h
kayak/flights-search          travel.transportation.search   verified · 3h
02

Pull and run

Fetch executes the schema on your machine. No hosted runtime in the loop — your CLI (or curl) talks to the site directly.

$ hermai fetch booking.com --action search \
    --param destination="Tokyo" \
    --param check_in=2026-05-01 \
    --param guests=2
03

Get structured JSON

Response comes back as clean JSON — the fields you saw on the catalog card, with real values. Ready for an LLM to reason over.

{
  "hotels": [
    {
      "name": "Park Hyatt Tokyo",
      "price_total": 842,
      "currency": "USD",
      "review_score": 9.2
    }
  ]
}

Benchmarks

Real sites. Real numbers.

Four sites where generic scrapers struggle, measured head-to-head against two tools agent builders actually reach for first: Firecrawl, the popular hosted scraper, and Claude Code's built-in WebFetch tool. Same network, cold fetch, single connection.

zillow.com favicon
Zillowzillow.com

Search for-sale listings in a ZIP

Response time
1.04s8.4×
8.80s
blocked
Agent tokens
2.2k3.3×
7.1k
blocked
Data quality
hermai

41 listings · 50 fields each

firecrawl

~9 listings in noisy markdown

webfetch

403 — PerimeterX wall

shopify.com favicon
Shopifyshopify.com

Fetch a product with its variants

Response time
792ms1.4×
1.12s
1.42s
Agent tokens
2.8k20×
57.0k
9.8k
Data quality
hermai

Full product · 7 variants · SKUs

firecrawl

1 product, mixed with promo HTML

webfetch

No inventory / no variant IDs

finance.yahoo.com favicon
Yahoo Financefinance.yahoo.com

Fetch full quote for a ticker

Response time
1.03s2.0×
2.11s
blocked
Agent tokens
18.7k
14.2k
blocked
Data quality
hermai

Full quote · 50 fields

firecrawl

Quote, but mixed with ads

webfetch

JS-rendered — empty payload

github.com favicon
GitHubgithub.com

Fetch repo metadata

Response time
776ms
844ms
612ms
Agent tokens
1.9k3.8×
7.1k
2.4k
Data quality
hermai

Repo metadata · 80 fields

firecrawl

README + UI chrome

webfetch

Summary only, rounded counts

April 2026 · Oakland, CA · cold fetch, single connection