Loading

Everything retrieved starts as a file. The Document and Image classes (plus Audio and Video) build files from every source you have, put() stores them with the provider once, and fromId() references the stored copy later without re-uploading.

Putting files in

use Crustum\Ai\Files\Document;

$stored = Document::fromString('CakePHP is a PHP framework.', 'text/plain')->put();

echo $stored->id . PHP_EOL;

fake_file_685b6577729e3e4c1287ad1eb6bdbee6

Every reader ends at the same put():

use Crustum\Ai\Enums\Lab;
use Crustum\Ai\Files\Document;

$stored = Document::fromPath('/var/www/knowledge.txt')
    ->withProviderOptions(['purpose' => 'assistants'])
    ->put(provider: 'openai');

echo $stored->id . PHP_EOL;

fake_file_2c8b70838a6eaf4b6dadec6b38eddc75

Readers: fromPath(), fromStorage() (filesystem disk), fromUrl(), fromString(), fromUpload(). Options set inline ride along (purpose), or branch per provider with a closure receiving the current Lab. Stored files attach to prompts via Document::fromId($stored->id). See Vector Stores for indexing them instead.

Reading and removing

use Crustum\Ai\Files\Document;

$file = Document::fromId('file-id')->get();

echo $file->mimeType() . PHP_EOL;

Document::fromId('file-id')->delete();

text/plain

get() returns the stored metadata; delete() drops the provider copy. Removing a file from a store keeps the stored copy unless asked. Re-adding it later is one add() call, not a re-upload.

See also

• Embedding — files to vectors
• Stores — indexing files for search
• RAG — the retrieval loop