ToolRobinNo account needed

Local Document Search

Find relevant passages in English PDF, TXT or Markdown files using a local MiniLM embedding model.

Choose up to 10 documents, then build the index. Enter a short English question. Review the five highest-ranked passages and their filenames/page numbers. Change the question to search again, or clear files to discard the in-memory index.

Drop files here or choose below

Self-made sample files. Processing still runs locally.

PDF / TXT / MD · up to 10 files · 5 MiB per file · Total 20 MiB · 1–100 PDF pages · 200 passages · 250,000 extracted characters. First-use model download, plus engine: 22.59 MiB. English documents and English questions; no generated answers. TXT and Markdown must use UTF-8. Passages are split to fit the local model. Dense technical text may reach the 200-passage limit sooner; split the files if needed. Questions must also fit the model: shorten a rejected question and retry without rebuilding the index.

    The model can miss a useful passage or rank an unrelated one highly. English is the supported retrieval language even when the interface is translated. Nothing is saved between page visits; copy needed source text before clearing.

    Help and limits

    What does this tool do?

    Find relevant passages in English PDF, TXT or Markdown files using a local MiniLM embedding model. Ask a question in different words and review matching source passages. It retrieves text; it does not write answers or decide whether a statement is true.

    Step by step

    1. Choose up to 10 documents, then build the index.
    2. Enter a short English question. Review the five highest-ranked passages and their filenames/page numbers.
    3. Change the question to search again, or clear files to discard the in-memory index.

    Choosing the settings

    OptionWhat changes
    Options and tradeoffsThe model is designed for English. Use specific questions; relevance is cosine similarity, not confidence. PDF page references are available; TXT and Markdown have filenames without invented page numbers.
    Formats and limitsPDF / TXT / MD · up to 10 files · 5 MiB per file · Total 20 MiB · 1–100 PDF pages · 200 passages · 250,000 extracted characters. First-use model download, plus engine: 22.59 MiB. English documents and English questions; no generated answers. TXT and Markdown must use UTF-8. Passages are split to fit the local model. Dense technical text may reach the 200-passage limit sooner; split the files if needed. Questions must also fit the model: shorten a rejected question and retry without rebuilding the index.

    Example and expected result

    If a document says “Returns are accepted within 30 days”, a question such as “How long do I have to send an item back?” can retrieve that passage without matching the same words. The score and rank depend on the corpus.

    If it is not working

    • For scanned PDFs, first extract text with OCR and save TXT. Split a corpus that exceeds the passage or character cap. A failed or cancelled index must be rebuilt before searching.
    • The model can miss a useful passage or rank an unrelated one highly. English is the supported retrieval language even when the interface is translated. Nothing is saved between page visits; copy needed source text before clearing.

    Formats and limits

    PDF / TXT / MD · up to 10 files · 5 MiB per file · Total 20 MiB · 1–100 PDF pages · 200 passages · 250,000 extracted characters. First-use model download, plus engine: 22.59 MiB. English documents and English questions; no generated answers. TXT and Markdown must use UTF-8. Passages are split to fit the local model. Dense technical text may reach the 200-passage limit sooner; split the files if needed. Questions must also fit the model: shorten a rejected question and retry without rebuilding the index.

    The model can miss a useful passage or rank an unrelated one highly. English is the supported retrieval language even when the interface is translated. Nothing is saved between page visits; copy needed source text before clearing.

    FAQ

    What does it do?

    Find relevant passages in English PDF, TXT or Markdown files using a local MiniLM embedding model. Ask a question in different words and review matching source passages. It retrieves text; it does not write answers or decide whether a statement is true.

    Which option should I choose?

    The model is designed for English. Use specific questions; relevance is cosine similarity, not confidence. PDF page references are available; TXT and Markdown have filenames without invented page numbers.

    Why is it not working?

    For scanned PDFs, first extract text with OCR and save TXT. Split a corpus that exceeds the passage or character cap. A failed or cancelled index must be rebuilt before searching.

    What should I check in the result?

    The model can miss a useful passage or rank an unrelated one highly. English is the supported retrieval language even when the interface is translated. Nothing is saved between page visits; copy needed source text before clearing.

    Is my file uploaded?

    No. Your files and text are processed locally in this browser, without uploading your input. Engines and models download when needed. Clearing or leaving the page discards the current session; download or copy any results you want to keep.

    You might also need