Explainers · measured July 2026
The Star Trek computer answered instantly. We built the layer that makes that real.
On the bridge, someone says something like “Computer, locate Commander Data,” and the answer arrives before the sentence finishes cooling. No spinner. No “please wait.” Every single time. It was set dressing, but it quietly promised something real: an assistant is only as fast as the layer that finds things for it. On a real 4.47-million-file machine, that layer is the difference between 67 seconds and 139 microseconds, and it is exactly what lets an AI agent answer like the ship’s computer instead of grinding grep.
TL;DR · the whole thing in five lines
Fiction’s starship computer answered “Computer, locate…” instantly, every time. That only works if the retrieval layer under it never makes you wait.
On a real 4.47-million-file machine, today’s retrieval layer does make you wait: Windows Search ~67 s, VS Code ~35 s, ripgrep 93.8 s per query.
Interlinked answers the same lookups in 139 microseconds (median of 20), ~480,000× faster than Windows Search on the same drive. A coffee refill versus a blink.
For an AI agent, that is the gap between 6 m 57 s and 71 tool calls versus 16 ms and one call (~7,200,000×) to find one file.
No AI in the box: a warm, always-fresh index (~1 ms behind your last save) that runs offline, sits near 44 MB at rest, and serves 19 AI clients from one install.
What the fiction actually promised
The one honest tell was the word “Working.”
A retrieval layer is the part of any assistant (fictional or real) that turns a request into a found answer: it locates the file, reads the content, and hands it back. Star Trek’s computer had the fastest one ever filmed. Its only concession to reality was a stock reply: “Working.” And then, a beat later, the answer.
On your machine, that “Working…” is a spinner that lasts a full minute. Ask Windows Search to find one file across a real 4.47-million-file drive and the median is 67 seconds, long enough to refill your coffee. Spotlight on a Mac is the same shape of promise with the same shape of wait; we cover that head-to-head in Spotlight vs Windows Search. The reason is not laziness: those tools re-walk the disk on every query, because nothing warmer exists machine-wide.
This is a metaphor, not a benchmark: Star Trek didn’t ship a spec sheet. But the interaction it imagined is a real engineering target, and the gap between the promise and your drive is measurable to the microsecond. Here is the layer that closes it.
“Computer, find X” · drawn honestly
Instant is not magic. It’s a warm index that never sleeps.
The ship’s computer never re-read the archive when you asked: it already held it, ready. That is the whole trick, and it is buildable today: keep an always-fresh index resident in the background, and a spoken request becomes a lookup, not an excavation. The command takes the top lane. Everything else takes the bottom one.
Same machine · same drive · same lookups
“Computer, locate…” priced out on a real bridge.
| The tool | The job | It takes | Interlinked | Faster by |
|---|---|---|---|---|
| Interlinked | one warm-index lookup | 139 μs | the baseline | |
| VS Code (Ctrl+P) | open one file you can name | 35 s | 85 μs | 412,000× |
| Windows Search | median of 20 file lookups | 67 s | 139 μs | ~480,000× |
| ripgrep (your agent's default) | one content query, same corpus | 93.8 s | 16 ms | 58,625× |
| Claude Code (no index) | find one file, agentically | 6 m 57 s | 16 ms | ~7,200,000× |
Methodology: Ryzen 9 9950X3D · 64 GB DDR5 · NVMe · Windows 11 · 4,470,000 files. Each row is self-consistent. The Interlinked column shows our time for that same job, so every multiplier is a like-for-like pairing: single-file lookups (VS Code, 85 μs), the median of 20 lookups (Windows Search, 139 μs), and content queries (ripgrep and the agent run, 16 ms). Full per-tool write-ups in the 412,000× VS Code piece and the ripgrep comparison (93.8 s → 16 ms on the same query, a minute and a half versus about the time it takes to say the word “Computer”).
One real agent · one real file
We asked a real agent to locate one file.
Six minutes and fifty-seven seconds of an agent grinding directory listings, versus a single blink, about 7,200,000×. The captain never watched the computer think out loud through seventy-one guesses; the answer just came. Tokens spent grepping are tokens not spent reasoning, and on a $200/month plan that is real money going to ls output the model forgets one turn later. We break the run down call-by-call in how Claude Code finds files.
The other half of the fantasy
The ship’s computer didn’t just find it. It knew how everything connected.
Ask it to “cross-reference” and it linked records the instant you finished the word. The same index that answers in microseconds also carries a knowledge vault: relationships it derives by itself (files that change together, repos that share hundreds of identical files, which project is alive and which went dormant) plus notes any agent can leave on any file that every other agent then reads. Keyed to each file’s identity on disk, so it survives renames, spans every repo, and belongs to no vendor. Not RAG, not your model’s memory: the layer under both, drawn out in the backbone of the LLM economy.
SPEED FINDS THE FILE. MEMORY KNOWS WHY IT MATTERS. THE COMPUTER HAD BOTH. SO DOES YOUR MACHINE NOW.
Questions people actually ask
The honest FAQ.
Did the Star Trek computer really answer instantly?
On screen, yes: the ship’s computer was written to respond the moment you asked, usually after a one-word “Working.” It is fiction, and this piece treats it as a metaphor, not a spec sheet. But the interaction it imagined is a real engineering target: an assistant is only as fast as the layer that finds things for it.
Is Interlinked actually as fast as “instant”?
Not literally instant, but past the point you can perceive: 139 microseconds median across 4.47M files on the cited machine. A blink lasts about 300 milliseconds: you could run roughly 2,000 of these lookups inside a single blink.
How is this different from Spotlight or Windows Search?
Same idea (you ask, it looks) but a very different result on a real drive. Spotlight on the Mac and Windows Search on the PC are the closest everyday things to the ship’s computer, yet a full-machine lookup takes tens of seconds to a minute. Interlinked answers in microseconds and hands the result straight to your AI agent, not just to you.
Is there an AI inside it?
No. No model, no embeddings, no cloud calls. Just a warm, always-fresh index plus a path/id match. It runs entirely offline and stays around 44 MB at rest, less than a browser tab.
Which AI tools can use it?
One install wires up 19 AI clients (Claude Code, Cursor, Copilot, Windsurf, Cline and more) over MCP. Everything local is free, forever: the app, whole-machine search, MCP. Hosting starts at $5.99/mo.
Give your agent the ship’s computer
Make “Computer, locate…” real on your machine.
Install once. One signed installer auto-configures 19 AI clients: Claude Code, Cursor, Copilot, Windsurf, Zed, Cline and the rest. No JSON editing.
Let it warm up. Filename search works immediately; content search fills in behind it. At rest the whole engine sits near 44 MB, less than a browser tab.
Then just ask. Whole-machine when you don’t know where a file lives, scoped to a folder when you do. A save is searchable about a millisecond later: the answer comes before you blink.
Everything local is free forever, on all your devices, no card. Hosting starts at $5.99/mo. Download for Windows.