Guides · finding a lost file · measured July 2026
You forgot the name. You didn't forget what's in it.
It is somewhere on your disk. You can picture a line from it (a number on an invoice, a name in the second paragraph, a sentence you typed at 1 a.m.), but the file itself? You saved it as scan-2019-invoice.pdf eighteen months ago and never read the name again. Guessing the name is hopeless. Searching for the thing you actually remember (a phrase from inside the file) is the move. Here is how, on any machine, why it crawls on Windows, and how to make it instant.
TL;DR · key takeaways
The short version. Everything below, condensed.
Stop guessing the filename. Search the words inside the file. The name is the one thing you forgot; a phrase you remember reading (a number, a name, a sentence) is what actually finds it.
Fast tools are blind; thorough tools are slow. grep and ripgrep read PDFs and Office documents as binary and skip them, and the built-ins that read them keep no warm index, so one content search on a full drive can take 93.8 seconds.
The fix is a warm, whole-drive content index: macOS Spotlight, but on Windows. Content search lands in 7 to 9 ms; the query ripgrep needs 93.8 s for comes back in 16 ms, 58,625× faster.
It reads inside PDFs and Office documents, stays current about 1 ms after you save, and idles near 44 MB, less than a single browser tab.
Your AI agents get the same index over MCP. A hunt that took Claude Code 6 minutes 57 seconds and 71 tool calls returns in 16 ms and one call. Everything local is free, forever: the app, whole-machine search, MCP. Hosting starts at $5.99/mo.
Do this first · every option, honestly
Search for what's inside, not what it's called.
Every one of these searches text inside your files, so a phrase you remember can find a file you can't name. They differ on two things that decide whether you get your answer in seconds or in minutes: whether they can read inside a PDF or an Office document at all, and whether they keep an index or re-read the disk on every query.
| The method | Where it is | Reads inside PDFs & Office docs? | Time · 4.47M files |
|---|---|---|---|
| Windows Search | Start menu · Explorer box | Yes, but only for files in its indexed locations | seconds → minutes |
| Explorer · “search file contents” | Any folder window | Text yes; PDFs & Office only with a filter installed | reads every file live, very slow |
| findstr · PowerShell | The command line | No: plain text only | grows with the drive |
| grep · ripgrep | The terminal · dev tool | No: reads PDFs & Office docs as binary, skips them | fast in one repo; walks the disk otherwise |
| Spotlight | macOS, built in | Yes, reads inside PDFs & Office docs | instant, but macOS only |
| Interlinked | Taskbar app + agent tools | Yes, including PDFs & Office documents | 7 to 9 ms, whole drive |
Methodology: Ryzen 9 9950X3D · 64 GB DDR5 · NVMe · Windows 11 · 4,470,000 files. The Interlinked figure is whole-drive content search across the machine (typically 7 to 9 ms). Command-line and Explorer times scale with the size of the drive, so they are given as ranges; the PDF and Office answers describe each tool's default behaviour on those file formats.
Why the reframe works
A name search reads the label. A content search reads the pages.
Here are three files on your disk: a spreadsheet, a meeting note, and a scanned invoice. On the left you do what most people do: type your best guess at the name. The search compares your guess to each file's label and matches nothing, because the label is exactly the part you forgot. On the right you type a phrase you remember reading. The search looks at the words inside each file, and one lights up.
That is the whole trick, and it is why the invoice is the hard case: it is a PDF. Its name is machine junk you never chose, and its text is locked inside a document format that plain-text tools skip. Search its contents and none of that matters. You asked for the one thing about it you actually kept: the number printed on the page.
The catch on Windows
The idea is easy. Doing it fast is not.
Two problems, not one
Coverage, then the walk.
Blindness. The tools that are fast (the command line, grep, ripgrep) only read plain text. A PDF or a Word document is a compressed container, so those tools see binary and skip the file entirely. Your phrase is right there on the page, and the search walks straight past it. Windows Search can read inside Office documents and PDFs, but only for files that sit inside its indexed locations, which on a big drive is a minority of them.
The walk. The tools that read everything have no standing index, so every query re-opens and re-reads files across the disk from cold. That work scales with the size of the drive, not the difficulty of your question, which is how a single content search on a full machine reaches 93.8 seconds. The fix for both is the same: index the contents of every file once, keep the index warm, and never walk again.
The anchor · macOS
Mac users already do this. It's called Spotlight.
On a Mac you type a phrase and the file appears (PDF, Pages, Word, all of it) because Spotlight keeps a whole-volume index of file contents and reads inside documents as it builds it. Content search feeling instant isn't exotic; it is what a warm, complete content index does. The catch is that Spotlight is macOS-only, and it was built for a person typing into a box, not for an AI agent making hundreds of lookups a minute. On Windows, and for agents, the same trick has to come from somewhere else.
Windows
Reads inside docs, but only in indexed locations
seconds → minutes on a big drive
macOS
Whole-volume content index, PDFs & Office included
as fast as you can type, Mac only
Windows
Whole-drive contents, PDFs & Office, always warm
7 to 9 ms · built for agents too
Spotlight proves the model: cover the whole volume, read inside the documents, keep it current. Reproduce that on Windows (and hand the same index to whatever is doing the searching, human or agent), and finding a file by a phrase you half-remember stops being a chore and becomes a reflex.
The fast answer on Windows
A warm content index. Every file, PDFs and docs included.
Interlinked keeps a live index of the contents of every file on the machine, and it reads inside PDFs and Office documents as it builds it, so a phrase you remember finds the file no matter what it is called or what format it is in. A content search across the whole drive lands in the 7 to 9 millisecond range (about 16 ms for an exact phrase). Edit a document and its new text is searchable about a millisecond later (under 30 milliseconds in the worst case). At rest the whole engine sits around 44 MB, and the index on disk stays under 1% of the drive.
It isn't only for you. The same index answers any AI client that speaks MCP: one install auto-configures 19 of them (Claude Code, Cursor, Codex, Cline, Continue and more). Agents are the worst case for the old way: the task that took Claude Code 6 minutes 57 seconds and 71 tool calls of hunting comes back in 16 milliseconds and one call, and the share of session tokens burned on finding files drops from roughly 58% toward zero.
Do this today
Find the file you can't name.
Install once. One signed installer builds the whole-drive content index and auto-configures 19 AI clients: Claude Code, Cursor, Copilot, Codex, Windsurf, Zed, Cline and the rest. No indexing settings to tune, no JSON to edit.
Search what you remember. Type the phrase you can still picture (a number, a name, a sentence) instead of guessing the filename. PDFs and Office documents are read inside, so the format never stops you.
Trust that it's current. Edit or add a file and its contents are searchable about a millisecond later. You never re-index, and no query falls back to re-reading the disk.
Let your agent use it too. The same content index answers Claude Code, Cursor, and Codex over MCP, so the file you found by a phrase, your agent can find the same way, in one call.
Everything local is free forever.
Desktop search plus agent access (the same content index handed to every AI client over MCP), free forever, on all your devices, no card. Hosting your projects starts at $5.99/mo. Download for Windows.
FAQ
The questions people actually ask.
How do I find a file when I've forgotten its name?
Search for a phrase you remember reading inside it (a number, a name, a sentence) instead of guessing the filename. A content search reads the words inside every file and surfaces the one that contains your phrase, even a scanned PDF you never named.
Can I search the text inside PDFs and Word documents?
Yes, but not with every tool. Fast plain-text tools like grep and ripgrep read PDFs and Office files as binary and skip them, so you need a search that extracts and indexes document text. Interlinked reads inside PDFs and Office documents as it builds its index, so the format never hides your phrase.
Why is searching file contents on Windows so slow?
Two separate problems. The fast tools are blind to documents, and the tools that read everything keep no standing index, so every query re-reads files across the disk from cold. On a full 4.47M-file machine that walk can reach 93.8 seconds; a warm content index answers the same query in 16 milliseconds.
Can my AI agent (Claude Code, Cursor, Codex) use the same search?
Yes, over MCP. One install auto-configures 19 AI clients, and they all query the same warm index. A file hunt that took Claude Code 6 minutes 57 seconds and 71 tool calls comes back in 16 milliseconds and a single call.
What does it cost?
Everything (including agent access over MCP, the same index handed to every AI client) is free forever, on all your devices, no card. Hosting starts at $5.99/mo.