Interlinked.
Download for Windows
Local · Private · Encrypted
Free forever. For you and your AI.
Works with Claude · Cursor · Codex · +16 more

Desktop AI, 1000× faster.

Drop-in replacement for the [search] inside every AI agent.

Interlinked
Precomputed into memory. Compressed to fit your whole machine.
8 ms
ripgrep
VS Code’s inefficient algorithm. 50× the calls.
17 min
ClaudeClaudeAnthropicDADario AmodeiDario AmodeiCodexCodexOpenAISASam AltmanSam AltmanVS CodeVS CodeMicrosoftSNSatya NadellaSatya NadellaCursorCursorxAIEMElon MuskElon Musk
Agent-agnostic.
ONE INSTALL
MCP
→
+ 15 more: 19 agents, zero config

Related files, surfaced on every search.

Your Search
“invoice”
invoice.pdf
found · 8 ms
contract.pdf
opened together · 6×
job photos
saved together
Your AI
the file + 2 related

Cross-repo work, unlocked. From any folder.

×Your agent today
One repo at a time: the folder it's launched in.
✓With Interlinked
Every repo on the machine: one session.
Where your AI agent’s tokens go
Finding the code · 58%
Thinking 28%
14%
71 calls
per query · ~7 min of grep
~$100
/mo burned on search
↑ frontier rates, only climbing

Interlinked finds it in 8 ms, one call, so you stop paying frontier rates to grep. Free, forever.

Encrypted: what it learns stays sealed

Everything Interlinked Files saves and everything it learns is encrypted at rest on your PC. The knowledge vault (notes and relationships on your files) is sealed with AES-256-GCM, with a master key wrapped by your machine’s TPM, the same hardware anchor BitLocker uses. Notes, clipboard history, and captures are stored in a SQLCipher-encrypted database with per-blob AES-256-GCM. Private: only this PC can read it. Encrypted: keys are sealed in your PC’s own security chip and never leave it. No cloud: no content telemetry, and search works fully offline. Nothing leaves your machine.

With Interlinked
Less than a second
one MCP call
Without Interlinked
6 min 57 sec
71 tool calls · sped up 100×
Works withWindows 10 & 11x64 · NTFSMCP▶▶
Private by construction

No cloud.

Files never leave your PC.

Just ask.

Any file. Under a second.

Don’t have an AI? Use it yourself.

interlinked
The desktop app

Already have one? Connect it.

Claude, Cursor, Codex
Private by construction

No cloud.
Files never leave your PC.

Just ask. Any file. Under a second.

Don’t have an AI? Use it yourself.

interlinkedThe desktop app

Already have one? Connect it.

+16 more
How we built it
2004
Windows Search was written. It has never been rebuilt since.
Still the default on every PC
2026
Rebuilt from zero

We started from zero. Benchmarked every modern search algorithm we could find, then rewrote the fastest ones from scratch: 50,000 lines, compiled to a native binary. No interpreter, no framework, no network: nothing stands between your question and the answer but the machine itself. That’s what milliseconds cost to build.

You’re meeting the Henderson team at 9 tomorrow.

Three years of proposals, contracts, invoices, and exported email threads across four folders. No file is named “Henderson”. The word lives inside them: PDFs, plain text, code.

Just say what you need“the contracts”
interlinked
henderson contract34 results · 6 ms
MSA-2023-signed.pdf
…this Henderson contract supersedes the 2022 MSA…
NDA-mutual-2022.pdf
…signed alongside the Henderson contract…
sow-amendment-notes.md
…Henderson contract term extended to 2027…
renewal-notes.txt
…dana: Henderson contract renewal attached…
+ 30 more
Ask your AI
“what do we have on Henderson?”
iFileText “henderson”  →  127 matches · 8 ms
Three proposals, the signed MSA, and two invoices. The March one is still unpaid.
The old way

Opening folders, file by file, until 11 p.m.

The new way

Type the name once. The match is inside the files: PDFs, Office docs, exports.

Home by 5.
every Henderson file, one search

One call · the round trip

It doesn’t just find the file. It hands back what moves with it.

Interlinked maps every file on the machine and how they connect. Your agent makes one MCP call and gets the ranked files and their relationships back: which files change together, what ships with what, plus any notes left on them.

INTERLINKED
MCP
Find files
Find relationships
Claude CodeCursorCodexCopilotWindsurf
1match
2match
3match
4match
the top 4 of
every file
on your machine
searched by name, content & document text
edited together
same repo
ships together

The full loop

1
You ask in your own words
“pull up everything from the Meridian deal”
2
Your AI searches once
search: meridian
3
Found in milliseconds, top 4
Meridian-MSA-signed.pdf
Meridian-onboarding.docx
meridian-renewal-notes.txt
Q3-proposal-Meridian.pdf
4Knowledge comes attached
Clients \ Meridian
closed Aug 2025, renewal due Q3
Meridian-MSA-signed.pdf
⟷ changes-with
final-quote-v3.xlsx· worked on together ×12: never named “Meridian”; only the link finds it
⟷ related
Meridian-onboarding.docx· same deal, lives in Legal\
⟷ saved-together
kickoff-thread.eml· same day, same folder
Q3-proposal-Meridian.pdf
⟷ related
pricing-model.xlsx· quoted in the proposal
⟷ opened-together
brand-deck-2024.pdf· reviewed side by side
+ 14 more connections…
Attached to the result, read by you and your AI

Scattered across the machine (some never even named for the deal), assembled and explained in one ask.

The expensive part isn’t the fix:
it’s finding what to fix.

Typical agent session  ·  ~50,000 tokens
FINDING THE CODE · 58%
■  Finding■  Thinking · 28%■  Answering · 14%
Claude Max · $200/mo · ~50% to code search
With Interlinked
0.001s
1 tool call
$100/mo
Saved
Without Interlinked
6m 57s
71 tool calls
~$100/mo
Burned on file hunting

Your AI assistant burns 71 tool calls searching for a single file. Interlinked gives it the answer in one call. Zero waste.

With Interlinked

Two ways an AI fails on your files.
Interlinked kills both.

Without it, your agent either burns minutes grinding through tool calls, or invents a wrong answer and hands it back with confidence. You pay tokens either way. Interlinked gives the agent the real file in one call, every time.

71 calls
1 call
Tool calls to find one file
7 minutes
8 ms
Wall-clock spent hunting
~58% of tokens
~0%
Context spent searching, not solving
Best guess
The real file
What the answer is built on
Attention on the hunt
Attention on the fix
Where the model's focus goes
Every file it opens
Only the one you need
What reaches the cloud

And the cost isn’t only time and tokens. Every wasted call packs the context with search noise, so by the time the agent finally has the file it has less room and less attention left for the thing you actually asked. Cut the hunt and you don’t just pay less: the answer gets better.

See it for real ·  same query, same machine

With Interlinked
Works with Claude, Cursor, Codex, VS Code
Less than a second
Without Interlinked: Slow
Claude Code · video sped up 100×
6 min 57 sec
Without Interlinked: Wrong
Confidently wrong · sped up 6×
41 sec · failed
How we built it
2004
Windows Search was written. It has never been rebuilt since.
Still the default on every PC
2026
Rebuilt from zero

We started from zero. Benchmarked every modern search algorithm we could find, then rewrote the fastest ones from scratch: 50,000 lines, compiled to a native binary. No interpreter, no framework, no network: nothing stands between your question and the answer but the machine itself. That’s what milliseconds cost to build.

Compatibility
One install.
19
AI assistants connected.

One background process. Every tool reads the same index. No per-tool setup.

Claude CodeCursorCodexCopilotWindsurfClineRoo CodeGemini CLIOpenClawZedContinueVisual Studio
One line, any clientclaude mcp add interlinked -- interlinked-mcp

+ 7 more, auto-registered by the installer

One engine. Four ways in.

Desktop, CLI, MCP, or the raw pipe: the same index answers them all. And every operator composes freely in a single query.

Access methods
Desktop App
search bar
12 ms
ifile CLI
shell & scripts
8 ms
MCP API
AI agents & LLMs
for agents0.8 ms
Direct pipe
your own client
0.2 ms

Median find-by-name latency · same index behind all four

In practice
YOU SAY

“The big videos I saved this month, biggest first.”

YOUR AGENT COMPOSES
size:>100mbtype:videomod:30dsort:size
conference-4k.mp42.4 GB
screen-demo.mov810 MB
drone-b-roll.mp4640 MB

Nothing to memorize: your agent writes the operators from the sentence. Answered in milliseconds.

Every operator, filter and sort: open the docs →

The engine your AI searches with

127,500× faster
17 min 13 s → 8 ms
One content search, whole machine.
  • ripgrep is the engine inside VS Code, the same one agents like Claude Code run.
  • It re-reads your whole disk on every search.
  • We replaced the scan with an index that’s already warm.

Same query, identical results.

2.35M files · methodology · 465 timed queries, 0 errors · the full table · 170+ measured write-ups in the research library

Everything local is free forever. Hosting from $5.99/mo.
Download for Windows
“find the invoice, and everything that moves with it”
invoice.pdf8 ms
opened with contract.pdf · 6×
saved with job photos · same job
note “client pays net-45, don’t chase early”