Search inside your documents by meaning, not just by words.
DocSeek lets you load a PDF, Word, HTML, Markdown or text file and find passages by what they mean — so you can locate the right paragraph even when you don't know the exact wording used.
A small AI model runs inside your browser to power the search. Nothing is ever uploaded.
DocSeek answers a very specific question: “Is what I'm looking for in this document, and where?” It reads your file, breaks it into short passages, and turns each passage into a numerical fingerprint (an embedding) using a compact AI model that runs entirely in your browser. When you type a query, it is fingerprinted the same way, and DocSeek ranks the passages by how close their meaning is to your query. Because it compares meaning rather than letters, it can surface a relevant paragraph even when it uses completely different words than you did. Everything — the file, the index, and your searches — stays on your device. No server, no cloud, no upload.
DocSeek extracts the text and reports how many sections and passages it found.
Click ⚙️ Build search index. The first time ever, a ~25 MB AI model
(all-MiniLM-L6-v2) downloads from a CDN and is cached by your browser; after that it works offline.
Indexing then fingerprints every passage — a progress bar shows how far along it is. The index is cached locally,
so re-opening the same file is instant.
Type what you're looking for and press Search. Results appear as passages, each tagged with its location (page number for PDFs, paragraph number otherwise), a match score, and the most relevant sentence highlighted.
| Mode | What it does | When to use it |
|---|---|---|
| Meaning search | Ranks passages by conceptual similarity using the AI index. | When you don't know the exact wording, or want everything about a topic. |
| Exact / keyword | Finds literal text, with an optional regex toggle. No AI index required. |
When you know the precise term, or want a quick literal cross-check. |
| Element | Meaning |
|---|---|
Location chip (p. 4, ¶ 12) | Where the passage sits in the document. |
Match score (72% match) | How close the passage's meaning is to your query. Higher is stronger; green ≥ 50%, amber ≥ 35%. |
| Highlight | In meaning mode, the single most relevant sentence; in keyword mode, every literal match. |
A modest top score (say 30–45%) doesn't mean failure — it often means the topic is only touched on lightly. If even the best matches look unrelated, that's good evidence the document doesn't cover what you asked.
Open 📋 Document overview and click Generate overview to get an extractive summary: DocSeek picks the most central real sentences from the document and lists them in reading order. Nothing is paraphrased or generated, so the overview is always faithful to the source. Choose how many sentences you'd like (3–20).
| Setting | Effect |
|---|---|
| Passage size | How many characters each passage holds. Smaller = more precise locations; larger = more context per result. |
| Overlap | How much text is shared between neighbouring passages, so ideas spanning a boundary aren't missed. |
| Results shown | How many passages to display per search. |
Changing passage size or overlap requires rebuilding the index (DocSeek will prompt you).
DocSeek processes everything locally in your browser. Your documents are never uploaded — not their text, not your
searches. The AI model and the supporting libraries are downloaded once from public CDNs and then cached; from
then on the tool runs offline. Your settings live in localStorage and the search index is cached in
IndexedDB, both on your own device.
.pdf · .docx · .html · .md · .txt
Technologies used:
Built with: