Comparisons · measured July 2026
Satya Nadella called it the age of copilots. Copilots still cannot find your files.
He is not wrong: the copilot is the interface of this decade, and Microsoft put one in every app it ships. But a copilot is only ever as good as the context it can hand its model, and that context lives in files. On a real 4.47-million-file machine we timed the layer underneath every copilot, the part that actually goes and gets the file. Windows Search took 67 seconds. A live whole-machine index took 139 microseconds. Same drive. Same file.
The short version · TL;DR
- 01Satya Nadella called it the age of copilots, a fair call. The copilot is the interface of the decade; the model is the brain.
- 02But every copilot answer begins by finding a file, and that layer is slow. On a 4.47M-file drive, Windows Search takes 67 seconds to find one file; VS Code's Ctrl+P takes 35.
- 03Both are Microsoft's own tools. The company that named the age of copilots still ships the slow floor they stand on.
- 04A live, always-warm whole-machine index answers the same lookup in 139 microseconds, about 480,000× faster than Windows Search on the same drive.
- 05Interlinked is that layer: no AI in the box, offline, one install that wires 19 AI clients (Copilot, Claude Code, Cursor and the rest) to instant file and content search.
What a copilot actually stands on
A copilot is an AI assistant built into an app: it drafts, suggests, and answers using whatever context the app can hand its model. That last clause is the whole story. The reasoning is fast and it gets faster every quarter. The handoff (finding the right file on your actual disk and reading it) is the slow part nobody rebranded.
The copilot is fast. The floor it stands on is not.
Every answer travels down this stack and back. The layers on top got faster; the one at the bottom (go find the file, on a real disk, then read it) is where the wait actually lives. It is the backbone of the whole thing, and it is the least glamorous layer in the picture.
The reasoning layers improve on a schedule. The file layer at the bottom is the one that decides how long you wait, and until now, nobody rebuilt it machine-wide.
Same machine · same drive · same file
The Microsoft stack, timed end to end.
Two of these are Microsoft's own: Windows Search and VS Code. The company that named the age of copilots ships both, and both make an agent wait tens of seconds for one file that is already sitting on the disk.
| The searcher | The task | Time | vs Interlinked |
|---|---|---|---|
| Interlinked | one live-index lookup | 85 μs | baseline |
| Windows Search | Microsoft · median of 20 file lookups | 67 s | ~480,000× |
| VS Code · Ctrl+P | Microsoft · same file, same drive | 35 s | 412,000× |
| ripgrep | what the agent shells out to · one content query | 93.8 s | 58,625× |
| Claude Code, no index | the full file hunt, start to finish | 6 m 57 s | ~7,200,000× |
Methodology: Ryzen 9 9950X3D · 64 GB DDR5 · NVMe · Windows 11 · 4,470,000 files. Each multiplier compares that tool to Interlinked on the matching task: single-file lookups against our 85 μs (median 139 μs, best 2 μs), the content query against our 16 ms. Full per-tool method in the 412,000× VS Code write-up and why Windows Search is slow.
One real task, watched
We watched a copilot's agent find one file.
Six minutes and fifty-seven seconds is long enough to make a coffee. Sixteen milliseconds is faster than you can notice it happened. Counted end to end (the whole hunt of 71 tool calls against the single call that replaces it), that is roughly 7,200,000× less work for the same file. And on a $200-a-month agent plan, when more than half the tokens go to finding files, more than half the bill does too. The fix is not a smarter model. It is a faster floor, the argument in the backbone piece.
Why the floor stayed slow
A copilot indexes the project you opened. Your machine is bigger than that.
GitHub Copilot reads the code around your cursor, your open tabs, and (for Chat) one workspace index. That is the honest scope of every copilot: the folder in front of it. But your machine has a dozen repos that ship together, plus the 90% of the disk that was never in git: configs, documents, downloads, the PDF with the answer. An agent's questions don't respect folder boundaries; its search tools do. See exactly what Copilot can and can't reach in how GitHub Copilot finds context.
The same whole-machine index that answers in microseconds also carries a knowledge vault: relationships the engine works out by itself (files that change together, repos that share hundreds of identical files, which project is alive and which went quiet) plus notes any agent can leave on any file that every other agent then reads. Keyed to the file's identity on disk, so it survives renames and belongs to no vendor.
NO AI IN THE BOX · OFFLINE · THE INDEX NEVER LEAVES THE MACHINE.
Questions people ask
FAQ
Is Interlinked the same as Microsoft Copilot?
No. Copilot is the assistant that talks to you; Interlinked is the file-and-retrieval layer underneath it. It runs no model of its own: it hands whichever copilot you use (Copilot, Claude Code, Cursor) the right file, fast, then gets out of the way.
Why can't Windows Search or Copilot just be fast?
They rebuild or walk the disk on demand, and each understands only the project you opened. A live, always-warm whole-machine index answers the same lookup in about 139 microseconds instead of 67 seconds, because the work is already done before you ask.
Does it work with GitHub Copilot, Claude Code, and Cursor?
Yes. One install configures 19 AI clients over the standard MCP protocol (Copilot, Claude Code, Cursor, Codex, Windsurf, Cline, Zed and more) with no JSON editing.
Do my files leave the machine?
Not unless you publish a project or a link yourself. There is no AI in the box, and search uploads nothing: the index sits on your disk (under 1% of the drive), the background service idles around 44 MB, and every result is a local lookup.
What does it cost?
Everything local is free forever, on all your devices, no card. Hosting starts at $5.99/mo.
Do this today
Give your copilot the bedrock.
Install once. One signed installer auto-configures 19 AI clients (Copilot, Claude Code, Cursor, Codex, Windsurf, Zed, Cline and the rest) over the standard MCP protocol. No JSON editing.
Let it index. Filename search works immediately; content search fills in behind it. At rest the whole engine idles around 44 MB (less than a browser tab) and the index takes under 1% of the drive.
Point any agent at it. Whole-machine when you don't know where something lives; scoped to a folder when you do. Either way the answer comes back in the microsecond range, fresh within about a millisecond of your last save.
Let the vault grow. Day one it knows your git history. Every week after, it knows your machine better, automatically, with no model in the loop.
Everything local is free, forever: the app, whole-machine search, MCP. Hosting starts at $5.99/mo. Download for Windows and hand your copilot the file layer it never had. New here? What is an MCP server?