Comparisons · measured July 2026
Minority Report promised instant recall. Your file search is still buffering.
In 2002, Minority Report imagined the year 2054: Tom Cruise stands before a wall of glass and pulls any record out of the air with a wave of his hand (video, files, faces) the instant he reaches for it. No spinner. No Indexing… 34%. More than two decades on, the gestures sort of arrived: we swipe glass all day. The instant recall did not. Ask a real Windows machine to find one file across a full drive and it still spins for 67 seconds. The future the movie sold was never really about the hand-wave. It was about the answer being already found the moment you asked, and that part finally shipped. It just looks like a warm index, not a pair of gloves.
TL;DR · the 30-second version
- 01Minority Report (2002) sold gesture-fast, zero-wait recall of any record. The real bottleneck was never the gesture. It was retrieval latency, and that's the part that never shipped for two decades.
- 02On a full 4.47-million-file drive, Windows Search still takes a median of 67 seconds to find one file. VS Code's Ctrl+P: 35 seconds on the same disk.
- 03A prebuilt, always-warm index answers the same lookup in 139 microseconds, about 480,000× faster than Windows Search, gone before the screen redraws.
- 04For AI agents the gap is brutal: Claude Code once spent 6 minutes 57 seconds and 71 tool calls hunting one file. The indexed call returns in 16 milliseconds.
- 05No AI in the box: an index plus exact matching, offline and private. Free forever, no card.
The same request, three eras
One file. Fiction, reality, and the fix.
The movie's magic wasn't the choreography. It was that the record was already found. Real desktop search re-walks the disk on every ask. A warm index does what the film only mimed: it has the answer waiting.
Same machine · same drive · same file
Instant recall isn't magic. It's a warm index.
A warm index is a prebuilt, always-fresh map of every file's name and contents that answers without walking the disk. Here is the exact same lookup, everywhere it happens today: 35 seconds of watching a spinner versus a flicker your eye can't catch.
| The searcher | The task | Time | vs Interlinked |
|---|---|---|---|
| Interlinked (warm index) | one indexed lookup | 85 μs | baseline |
| VS Code · Ctrl+P | same file, same drive | 35 s | 412,000× |
| Windows Search | median of 20 queries | 67 s | ~480,000× |
| ripgrep (your agent's grep) | same content query | 93.8 s | 58,625× |
| Claude Code (no index) | find one file, agentically | 6 m 57 s | ~7,200,000× |
Methodology: Ryzen 9 9950X3D · 64 GB DDR5 · NVMe · Windows 11 · 4,470,000 files. Single-file lookups are the 85 μs best case; the machine-wide median across 20 queries is 139 μs. The ripgrep row is a content query over the same corpus; the Claude Code row is the measured agent run. Full per-tool method in the 412,000× VS Code write-up and the ripgrep comparison.
Why the future ran late
The movie copied the wrong thing.
Hollywood even hired a real interface researcher, John Underkoffler, to design that gesture wall, and he later shipped it for real as a spatial-computing product. Impressive, and beside the point. Moving windows with your hands is theater. The thing that made the scene feel like 2054 was that the record appeared with zero wait. The interface was solved. Retrieval never was.
Because every tool built for a person walks the disk on demand. VS Code indexes only the folder you opened. Windows Search consults a throttled, partial index and then trudges the filesystem, thirty years of good intentions that still add up to 67 seconds. And your agent shells out to ripgrep, re-reading the drive on every single query: 93.8 seconds where the index takes 16 milliseconds.
The fix is boring and total: keep an always-fresh index warm in the background so the answer is already found when you ask. 139 microseconds, machine-wide. That is the whole trick the precogs were standing in for.
One real task, measured
Your agent already does the hand-wave. For six minutes.
Six minutes fifty-seven seconds, 71 tool calls: an agent conducting your filesystem like Anderton conducting the precog feed, except blind, re-reading directories it forgets a turn later. Give it the index and the same hunt collapses to a single call; on the full workflow that's roughly seven-million-fold less waiting. Tokens spent grepping are tokens not spent thinking: the backbone layer nobody optimizes. Here's where an agent's day actually goes.
The part that actually shipped
No precogs. No gloves. Just the answer, already found.
139 microseconds is impossible to feel: a screen refreshes every ~16 milliseconds, so the index answers more than a hundred times inside a single frame (here's what a microsecond actually is). And unlike the precogs, it isn't guessing the future. It's reading a live index of your present: save a file and it's findable about a millisecond later, no reindex button, no waiting.
THE INSTANT-RECALL FUTURE ARRIVED QUIETLY, AS A BACKGROUND DAEMON, NOT A HAND-WAVE.
Frequently asked
Fiction, fact, and the fix.
Did Minority Report actually predict anything real?
The gestures, sort of. The film's hand-wave interface was designed by a real researcher, John Underkoffler, and later shipped as an actual spatial-computing product. But the part the story leaned on (a record appearing the instant you reach for it, with zero wait) stayed fiction. That zero-latency retrieval was always the hard part, and it's the part that finally shipped as a warm index rather than a pair of gloves.
Why is Windows Search still so slow?
On a full 4.47-million-file drive it consults a throttled, partial index and then falls back to trudging the filesystem: a median of 67 seconds across 20 real queries on our test machine. It was built for a person occasionally hunting a document, not for an agent firing hundreds of lookups a minute.
What makes Interlinked instant?
A prebuilt, always-fresh index of every filename and file's contents, kept warm in the background (around 44 MB at rest). Queries don't walk the disk. They hit the index. Median lookup is 139 microseconds across 4.47M files, and a saved edit becomes findable in about a millisecond.
Is there really no AI inside?
Correct. No model, no embeddings, no cloud. It's an index plus exact matching, running offline and private on your machine. That's precisely why it answers in microseconds instead of seconds: there's nothing to infer, only something to look up.
How do I give my AI agent this instant recall?
One signed installer wires the search up to 19 AI clients: Claude Code, Cursor, Copilot, Codex, Windsurf, Zed, Cline and the rest. Everything local is free, forever: the app, whole-machine search, MCP. Hosting starts at $5.99/mo.
Do this today
Give your agent instant recall.
Install once. One signed installer wires the search up to 19 AI clients: Claude Code, Cursor, Copilot, Codex, Windsurf, Zed, Cline and the rest. No JSON editing.
Let it warm up. Filenames are searchable immediately; contents fill in behind them. At rest the whole engine sits around 44 MB, less than a browser tab.
Save and forget. Edit a file and it's findable about a millisecond later. The index rides your changes. There is no reindex button to press.
Inherit day-one memory. On a fresh machine it read 8,455 git events across all 12 repos in 1.7 seconds and drew 26,958 relationships: context your agent gets for free.
Everything local is free, forever: the app, whole-machine search, MCP. Hosting starts at $5.99/mo. Download for Windows.