Postbuild: any framework or host

Seekite only needs rendered HTML and a directory that your host publishes. For frameworks without an adapter, run the ordinary CLI after the site build:

{
  "scripts": {
    "build": "your-framework build",
    "postbuild": "seekite build --output dist/search"
  }
}

Point generatedDir at the framework output in search.config.ts. You can narrow noisy layouts without writing a custom parser:

import { defineSearch, generatedHTML } from "@seekite/build"

export default defineSearch({
  generatedDir: "dist",
  output: "dist/search",
  corpora: {
    docs: {
      source: generatedHTML({
        bodySelector: ["[data-seekite-body]", "main", "article"],
        excludeSelectors: ["nav", "aside", ".table-of-contents"],
      }),
    },
  },
})

Build caches

Persist .seekite/cache between deploys. Seekite keys every row by provider ID and exact embedding input, so the outer CI cache key should stay stable.

For GitHub Pages:

- uses: actions/cache@v4
  with:
    path: .seekite/cache
    key: seekite-embeddings-${{ runner.os }}
    restore-keys: seekite-embeddings-
- run: pnpm build
- run: pnpm exec seekite build --output dist/search

On Netlify or Cloudflare Pages, point SEEKITE_CACHE_DIR at the provider's persistent build-cache directory, then use the same build and postbuild commands. Run seekite cache prune --max-size 500MB before indexing when the host imposes a cache quota.

Serve content-addressed corpus shards with Cache-Control: public, max-age=31536000, immutable. Serve manifest.json with Cache-Control: public, max-age=0, must-revalidate so a deploy switches to the new shard set immediately.