Files & search

#Files, type: storage

A collection declared "type": "storage" is R2-backed: each document is a file (blob) addressed by its path, with metadata fields you declare. Auth is exactly the same rules model as any collection, so file access is scoped by path just like data.

{
  "users/$userId/files/$fileId": {
    "type": "storage",
    "fields": { "name": "String", "owner": "String!", "size": "UInt" },
    "rules": {
      "read":   "@user.id != null && $userId == @user.id",
      "create": "@user.id != null && $userId == @user.id && @newData.owner == @user.id",
      "update": "false",
      "delete": "@user.id != null && $userId == @user.id"
    }
  }
}

• The path scopes the file. users/$userId/files/$fileId with $userId == @user.id means a user can only touch files under their own id. The path is the access boundary, enforced by the runtime rule.
• Declared fields are the file’s metadata (name, size, content-type, whatever you need); the bytes are stored separately in R2 and streamed.
• Mark owner (and any set-once metadata) ! so it can’t be reassigned.
• Storage collections are offchain.

Upload and download go through the SDK: setFile(path, file, { metadata }) uploads a File (or null to delete); getFiles(path) lists files under a path. The same path-scoped rules apply.

import { getCurrentUser, setFile, getFiles } from "@bounded-sh/client";

const user = getCurrentUser();
if (!user) throw new Error("Sign in first");

await setFile(`users/${user.id}/files/avatar`, file, {
  metadata: { name: file.name, owner: user.id },
});
const { data: files } = await getFiles(`users/${user.id}/files`);

#Search, search: { fields: [...] }

Declare which String fields are full-text indexed. The runtime maintains the index; you query it through the data plane.

{
  "orgs/$orgId/docs/$docId": {
    "fields": { "org": "String", "title": "String", "body": "String" },
    "tier": "durable",
    "search": { "fields": ["title", "body"] },
    "rules": {
      "read":   "@user.id != null",
      "create": "@user.id != null",
      "update": "@user.id != null",
      "delete": "@user.id != null"
    }
  }
}

• fields must be a non-empty array of valid field names.
• Index only the fields you actually search; each adds write-time index cost.
• Search respects read rules, results a caller can’t read are not returned.

Search is a query mode on the collection, combinable with filters and paging:

// SDK
import { search } from "@bounded-sh/client";
const hits = await search("orgs/o1/docs", "quarterly revenue", { fields: ["title", "body"] });

// CLI
// bounded data search --app-id <id> --path orgs/o1/docs --query "quarterly revenue"

#Choosing between them and ordinary data