GTC AI LAB · REAL MODELS, LOCAL PROCESSING

Your document. Your device.

Read text from an image or find a relevant passage in a document. The AI runs in your browser. You choose when to download a model, and your input stays on your device.

No account. No installation. No per-use AI fee.

English document processing · English and Arabic interface

TRY IT ON YOUR DEVICE

Two useful tasks. Real inference.

Start with a fictional sample, or use your own input. These tools extract and retrieve information; review the output before using it.

01 / OCREnglish

Read text from an image

Turn an English receipt, invoice photo or screenshot into selectable text. This reads text; it does not validate an invoice or identify its fields.

PP-OCRv6 Tiny · ~6.3 MB model + ~14.3 MB runtime, plus app files

Model not loaded
PNG, JPG or WebP · up to 5 MiB and 12 million pixels · English text
02 / SEARCHEnglish

Find a passage by meaning

Ask for an idea in different words. Local semantic search ranks the relevant excerpts from your English document. It does not generate an answer.

MiniLM-L6-v2 · ~23.7 MB model + ~26.9 MB runtime, plus app files

Model not loaded
English · up to 20,000 characters and 24 short passages · query up to 300 characters

Relevant passages

Matching excerpts will appear here.

    Similarity measures a semantic match. It is not a probability that a statement is correct.

    PROCESSING & STORAGE

    Your input stays here.

    Images, document text, queries and results are processed in this tab. They are not uploaded, logged or used to train a model. GTC serves the app and model files, so normal hosting request metadata is still generated. Closing or resetting the tool clears its current input and results.

    WebAssembly runs on your device. Loading and processing speed depend on memory, browser and hardware. If a model cannot run, the tool shows an error; your input is not sent to a cloud service.

    About the first download

    Models and runtime files are downloaded from GTC only after Load model. Your browser can cache model files for later visits. It may evict them or require another download; this is not a promise of fully offline access.

    Unload both models first. This removes only the Lab’s saved model caches. Your browser manages its ordinary HTTP cache separately.

    Open technology, clear limits

    Built with runonweb by midudev (MIT), PP-OCRv6 by PaddlePaddle and MiniLM by sentence-transformers (Apache-2.0). ONNX Runtime and Transformers.js run the models locally. The interface supports Arabic; these selected models are offered for English documents.

    A useful tool is the start of a workflow.

    Connect document intake, reviewed results and your business systems. Garrich Torres Consultants can scope the integrations, permissions and operating support around your team.

    Build this into your workflow