CanDoYa
All tools
Saved toolsYour favorites

Add to favorites

Saved in this browser only.
Private browser utility / Social & Creator

Free Question Answering Tool

English only

Runs entirely in your browser - no upload, no sign-up.

Live workspaceLocal processing

Paste English context and ask a factual question to extract a source-backed answer.

0 of 6000 characters

Include the fact in a focused passage. Longer text is read in overlapping sections.

0 of 300 characters

Ask for a short fact that appears directly in the context.

Your source-backed answer appears here

You will see the extracted phrase, a confidence score, and the surrounding source passage for verification.

Your context and question stay in this browser worker. Only the public AI runtime and model files are downloaded.

Share this tool
question answering / browser utility
01 / Overview

What is a question answering tool?

A question answering tool finds a short answer to a question inside context you provide. This extractive reader searches English text for one continuous answer span, reports its model confidence, and shows the source passage. The model runs inside your browser, so your context and question are not uploaded for inference.

02

How to use

  1. 01
    Paste the source context

    Enter up to 6,000 characters of focused English text that contains the fact you want to find.

  2. 02
    Ask a specific question

    Use a factual question whose answer should be one continuous phrase in the context.

  3. 03
    Run the local reader

    Select Find answer, then let the first model download and on-device reading pass finish.

  4. 04
    Check the evidence

    Review the highlighted source span and confidence score, then copy the result or refine the question.

03

Who it's for

  • Students and researchers locating a date, person, place, quantity, or definition in focused English notes.
  • Meeting participants checking transcript excerpts for a stated deadline, owner, decision, or follow-up.
  • Support and operations teams finding a specific policy detail in pasted documentation without uploading internal text for inference.
  • Editors and fact-checkers tracing a short answer back to the exact passage before quoting or summarizing it.

This tool uses Xenova's uncased DistilBERT model fine-tuned on SQuAD 1.1 through Transformers.js. It divides longer context into overlapping passages, reads each passage in a browser worker, and ranks the returned spans. WebGPU is tried first, with a WebAssembly CPU fallback when graphics acceleration is unavailable.

Extractive question answering does not write a new explanation or search the web. It selects words already present in the supplied context. The first run downloads public runtime, tokenizer, and model files that total tens of megabytes. Browsers normally cache them, but cleared storage or private browsing can cause another download.

FAQ

Is my context or question uploaded?

No. The context and question are sent only to a worker in your browser for inference. The worker downloads the public Transformers.js runtime and DistilBERT model files from their hosts, but those file requests do not include the text you entered. CanDoYa does not run a question answering API for this tool.

Is the question answering tool free?

Yes. There is no account, API key, payment, or server-side usage quota. Your device performs the model inference. You still use network data for the first public runtime and model download, and the browser needs enough memory and storage to load and cache those files.

How much context can I use?

You can enter up to 6,000 Unicode characters of context and a question up to 300 characters. Longer context is divided into overlapping passages of at most 1,400 characters. Focused context usually gives clearer results because fewer unrelated phrases compete as possible answer spans.

What kind of questions work best?

Ask factual English questions whose answer appears as one continuous phrase, such as who, when, where, how many, or what a term means. Questions that require arithmetic, outside knowledge, opinion, multi-step reasoning, or combining distant passages do not match this extractive model's training task.

What does the confidence score mean?

The score ranks how strongly the model preferred the selected start and end tokens for this question and context. It is not a guarantee that the answer is true or complete. Compare the highlighted passage with the question, especially when the score is low or the wording is ambiguous.

Can the model say that the context has no answer?

Not reliably. This DistilBERT checkpoint was fine-tuned on SQuAD 1.1, where each training question has an answer in the passage. It will often select some span even when your context does not contain a good answer. A weak or unrelated result should be treated as no useful answer and checked manually.

Which languages does the reader support?

The selected DistilBERT checkpoint and SQuAD fine-tuning are English, so the page marks the tool English only. Text in another language may still produce an output, but that output lacks the same training basis and should not be treated as a reliable reading-comprehension result.

Does question answering work without WebGPU?

Yes. The worker tries WebGPU when the browser and graphics adapter expose it. If that path is unavailable or fails, the tool restarts with the WebAssembly CPU backend. The fallback keeps the context on your device, but loading and inference may take longer.