Interlinked.

Comparisons · measured July 2026

Interlinked vs Windows Search.

You press the Windows key, type a filename, and wait. On a real 4.47-million-file PC, the search built into Windows takes about 67 seconds (median) to surface a file that lives outside its default index. The same lookup, on the same drive, through a whole-drive index comes back in 139 microseconds, while the engine sits at about 44 MB of memory. Mac users don't recognise this problem, because Spotlight answers as fast as they can type. Here is exactly why the gap is that wide, and why it comes down to coverage, not speed.

TL;DR · Key takeaways

Short version: it's coverage, not speed.

  • On a 4.47-million-file PC, the search built into Windows takes about 67 s (median) to surface a file outside its default index. The same lookup through a whole-drive index returns in 139 µs: 85 µs on a single file, roughly 480,000× faster.

  • It comes down to coverage, not speed: a partial index turns every miss into a slow filesystem walk. VS Code's picker outside its workspace is 412,000× slower on the same lookup; Spotlight feels instant only because macOS indexes the whole volume.

  • Content search across the whole drive lands in 7 to 9 ms: the query that took ripgrep 93.8 s comes back in 16 ms (58,625× faster), because the text is already indexed.

  • The worst case is an AI agent: a file hunt that took Claude Code 6m57s across 71 tool calls returns in 16 ms and one call (~7,200,000× faster), and the ~58% of session tokens spent hunting drop toward zero.

  • It stays invisible while it works: about 44 MB of memory at rest, ~1 ms to re-index a file you just saved. Everything is free forever, no card. Hosting starts at $5.99/mo.

The whole story in one picture

It isn't speed. It's what each one indexes.

Read this top to bottom. The strip on the left is what each tool keeps in its index; the bar on the right is how long one lookup takes, on a shared log scale. The two are linked. Index a slice of the drive and every miss becomes a slow walk of the rest. Index the whole drive and there is nothing left to walk. Coverage is the cause; the time is the effect.

WHAT IT INDEXESTIME TO FIND ONE FILE · LOG SCALEindexedwalked live100µs1ms10ms100ms1s10s100sLOG SCALE: EACH STEP ×10WINDOWS SEARCHthe built-in · Windowsa subset: profile folders, Start menu67 smedian · outside the indexSPOTLIGHTthe built-in · macOSWHOLE STARTUP VOLUMEindexed by default, kept currentAS FAST AS YOU TYPEmacOS only, not timed on this machineINTERLINKEDwhole-drive index · WindowsEVERY FILE + CONTENTS139 µsmedian of 20 · 85 µs on a single file

Windows Search covers a subset by default and hatched-walks the rest, so a file in the vast middle of the disk lands on a 67-second walk. Spotlight covers the whole startup volume, which is why it feels instant, but it is macOS-only. Interlinked covers every file and their contents on Windows, so the bar collapses to a direct lookup: 139 µs.

The head-to-head

Line by line, on the same machine.

The two Windows tools are timed on the exact same drive; Spotlight is the macOS anchor, cited for what it covers rather than a number measured here. Every difference below feeds the last row.

DimensionWindows SearchSpotlightInterlinked
PlatformWindowsmacOSWindows
What it indexesA subset by default: profile folders, Start menuThe whole startup volumeEvery file on the machine: names and contents
If the file is not indexedFalls back to walking the filesystemRarely happens: coverage is fullCannot happen: coverage is complete
FreshnessThrottled indexer; can lag behindKept current~1 ms after you save (under 30 ms worst case)
Idle memoryA throttled background serviceA background service~44 MB in Task Manager
Time to find one file67 s medianas fast as you type139 µs median

Methodology: Ryzen 9 9950X3D · 64 GB DDR5 · NVMe · Windows 11 · 4,470,000 files. The Windows Search figure is a median for a file that lives outside the default indexed folders; the Interlinked figure is the median filename lookup across 20 queries on the same corpus (85 µs on a single file, 2 µs best case). We don't assign Spotlight a millisecond number because it runs on a different OS. More on the built-in in why Windows Search is slow and how to search files on Windows.

Why a miss costs a minute

A miss doesn't fail. It walks.

Windows Search indexes a subset of paths by default: your profile folders, the Start menu, mail. Ask for a file inside that slice and it answers quickly. Ask for anything outside it (Program Files, a code repository, another drive, the download from last spring) and Windows falls back to walking the filesystem folder by folder, checking names as it goes. That walk scales with the size of the disk, not the difficulty of your question. On 4.47 million files, it is where the 67 seconds goes.

WINDOWS SEARCHQUERYCHECK A SUBSET INDEXmissWALK THE FILESYSTEM67 sINTERLINKEDQUERYWARM WHOLE-DRIVE INDEXhitDIRECT LOOKUP139 µsSame query. The only difference is whether the file was already in the index.
67 s
Windows Search median: a file outside the indexed folders, on 4.47M files
139 µs
same drive, same query, with every file already in the index
~480,000×
the gap between reading an answer and computing it from scratch

This isn't unique to Windows Search. Every tool that lacks a whole-drive index lands on the same wrong side of the same cliff: VS Code's file picker outside its workspace (412,000× slower on the same lookup), the command line, an AI agent shelling out to grep. Coverage is the dividing line, and a partial index is the same as no index for any file that falls outside it.

The anchor · macOS

Spotlight already proved the point.

The reason Mac users expect file search to be immediate is that macOS indexes the whole startup volume by default and keeps it current, so a search almost never falls back to a walk. That is the entire trick: not faster hardware, not smarter ranking. The index simply covers the files you ask for. Spotlight is proof that whole-volume coverage is normal, expected, and shippable.

Two things it can't do, though. It is macOS-only, so it doesn't help the Windows machine you actually run. And it was built for a human typing into a box, not for an AI agent making hundreds of file lookups a minute, and not for searching the full text inside every document and source file. On Windows, and for agents, that same coverage plus freshness has to come from somewhere else.

One warm index · about 44 MB

Spotlight's coverage, on Windows, for a browser tab.

Interlinked keeps a live index of every file on the machine (names and contents) and stays current as files change, so nothing ever falls back to a walk. A single filename lookup returns in about 85 microseconds (139 µs median across 20 queries). Content search across the whole drive lands in the 7 to 9 millisecond range. Save a file and its contents are searchable roughly a millisecond later. And it does all of it quietly: at rest the engine sits around 44 MB, and the index on disk stays under 1% of the drive (about 24 GB on a 4 TB machine).

139 µs
median filename lookup across 20 queries on 4.47M files
7 to 9 ms
content search across the whole drive: common words ~9 ms, rare symbols ~7 ms
~1 ms
from saving a file to its contents being searchable, under 30 ms worst case
44 MB
at rest in Task Manager, less than a single browser tab

The worst case for any file search is an AI agent (hundreds of lookups per session), so Interlinked hands the same index to any AI client that speaks MCP: one install auto-configures 19 of them (Claude Code, Cursor, Codex, Cline, Continue and more). Same agent, same query: the lookup that took Claude Code 6 minutes 57 seconds and 71 tool calls comes back in 16 milliseconds and one call, and the session tokens that used to go to hunting for files drop from roughly 58% toward zero.

Do this today

Give the search that's already on your PC nothing left to walk.

1

Install once. One signed installer builds the whole-drive index and auto-configures 19 AI clients: Claude Code, Cursor, Copilot, Codex, Windsurf, Zed, Cline and the rest. No indexing options to tune, no JSON to edit.

2

Let it cover everything. Filename search works immediately; content search fills in behind it. Because the index covers the whole drive, no query ever falls back to a walk: the exact thing that makes Windows Search slow simply can't happen.

3

Scope when you know where to look. Whole-machine when you don't know where a file lives; scoped to a folder when you do. Both answer in the microsecond-to-millisecond range: scoping just makes it more precise.

4

Search from the taskbar, or from your agent. Type in the app for yourself, or let Claude Code, Cursor and Codex query the same warm index over MCP. One index on the machine, every searcher on it.

Everything local is free forever.

The full product (the same index exposed 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.

Common questions

The questions people ask next.

Isn't Windows Search already indexing my drive?

It indexes a subset by default (profile folders, the Start menu, mail) and falls back to walking the filesystem for anything outside it. On a 4.47-million-file PC that walk is where the 67 s median goes; a whole-drive index has nothing left to walk, so the same lookup returns in 139 µs.

Will a whole-machine index slow down my PC?

No. At rest the engine sits around 44 MB in Task Manager (less than a single browser tab) and a file you save is searchable again about 1 ms later, so the index is never stale and never busy.

Can it search inside file contents, not just names?

Yes. Content search across the whole drive lands in the 7 to 9 ms range; the same query that took ripgrep 93.8 s comes back in 16 ms (58,625× faster), because the text is already indexed instead of scanned on demand.

How does this help an AI agent like Claude Code or Cursor?

Agents are the worst case for file search: hundreds of lookups a session. One install hands the same warm index to any AI client that speaks MCP, and a file hunt that took Claude Code 6m57s across 71 tool calls returns in 16 ms and one call (~7,200,000× faster), with the ~58% of session tokens spent hunting dropping toward zero.

What does it cost?

Everything is free forever, no card. Hosting starts at $5.99/mo.

Measured July 2026 · Ryzen 9 9950X3D · 64 GB DDR5 · 4.47M files · Windows 11← All research