Get started

ocd must be ready. The operator must configure OpenAI-compatible embedding (and optional rewrite / rerank / chat) providers in the platform config before AI Search can index or answer.

1. Create namespace and instance

Use pinned Wrangler against live ocd (commands follow Cloudflare AI Search / Wrangler resource flow).

2. Declare bindings

{
  "name": "search-app",
  "main": "src/index.ts",
  "ai_search_namespaces": [{ "binding": "SEARCH_NS", "namespace": "team" }],
  "ai_search": [{ "binding": "SEARCH", "instance_name": "docs" }],
  "ai": { "binding": "AI" }
}

bun run oc types --config wrangler.jsonc

3. Upload and search

export default {
  async fetch(request: Request, env: Env): Promise<Response> {
    if (request.method === "PUT") {
      const file = await request.blob();
      const item = await env.SEARCH.items.upload("guide.pdf", file);
      return Response.json(item);
    }
    return Response.json(await env.SEARCH.search({ query: "how does cache work?" }));
  },
} satisfies ExportedHandler<Env>;

Upload returns quickly with a queued item; parse / chunk / embed / index continue asynchronously on the node.

4. Markdown Conversion

const formats = await env.AI.supported();
const md = await env.AI.toMarkdown({ name: "page.html", blob: htmlBlob });

env.AI.run(), models(), gateway, AutoRAG, and other inference members are rejected.

5. Deploy

bun run oc deploy --config wrangler.jsonc

Next: Concepts.