Prism: image metasearch (Go + HTML)

Search Yandex, Pinterest or Danbooru from one clean UI. Standard library only: no dependencies to download.

Run

go run .                    # http://127.0.0.1:8080
go run . -addr :3000        # custom address (or ADDR=:3000)

Build

go build -o prism .         # single binary, UI is embedded
./build.sh                  # cross-compiles linux/mac/windows into ./dist

Needs Go 1.21 or newer.

How each engine works

• Yandex: reads the server-rendered results page and extracts the serp-item JSON blobs (same idea as 4get). Paged with p=. Yandex may show a captcha after many requests; the UI says so.
• Pinterest: the internal BaseSearchResource endpoint (as in Binternet), paged with Pinterest's bookmark token.
• Danbooru: public JSON API (/posts.json). Anonymous accounts can use 2 tags per search. "Safe results only" keeps ratings g and s.

/api/img?u=... is a small image proxy for downloads and hotlink fallbacks. It only returns image/* responses and refuses private/loopback addresses.

Files

main.go (server, /api/search), scraper.go (types, HTTP, cache), yandex.go, pinterest.go, danbooru.go, proxy.go, web/index.html (UI)

Troubleshooting

• Danbooru "no such host": your DNS blocks the domain. Prism retries with 1.1.1.1 / 8.8.8.8 / 9.9.9.9 (override with DNS_SERVERS), then DNS-over-HTTPS. If your ISP also blocks the site by SNI/IP, use a VPN or a mirror (DANBOORU_HOST=...).
• Danbooru 403: Cloudflare rejects browser-like user agents on the API; Prism sends an honest bot UA (botUA in scraper.go).