26×
less for your AI to read.
3,982 tokens per question instead of 105,132 reading the files whole, on 107 real questions.
A private search engine for your AI
Semlith
It reads the folders you point it at — code, notes, PDFs, slides — so your AI assistant can find the exact part it needs, in milliseconds, without anything leaving your machine.
Out of everything you have, it finds the part that matters.
That's the whole idea. Your AI stops reading everything — or guessing — and goes straight to the answer.
It reads Code, Notes, PDFs, Word docs, Slides, Spreadsheets, Notebooks, Web pages, Images, Markdown.
01 — The problem
AI assistants spend most of their time reading the wrong things.
Ask one a question about your project and it has two bad options. Semlith gives it a third.
Tokens are the little chunks of words an AI reads. Fewer tokens means faster and cheaper. Measured by asking 107 real questions about Semlith's own code.
02 — Meaning, not just matching
It understands what you mean — not just the words you typed.
Semlith cuts everything you give it into small pieces…
…and places each piece on a map, by what it's about.
Pieces that mean the same thing land side by side — even with no words in common. Your question lands right next to its answer. (Exact names still work too: it checks the words and the meaning, then combines both.)
03 — Connections
It also maps how everything connects.
Which piece uses which. So your AI can answer the questions search alone can't: what uses this? and what would break if I changed it?
Certain — the file says so
Confirmed — only one match
A few options — it says how many
A hint — worth a second look
Hover a piece →
Watch it light up everything it touches.
04 — Always up to date
You hit save.
Fix a bug, edit a note, drop in a new PDF.
Semlith notices.
It quietly watches in the background, using almost no power — under one second of work per idle minute.
It re-reads only what changed.
One file changed? One file re-read. Everything else stays exactly where it was.
Ready to be found.
From the moment you save to the moment your AI can find it. No rebuild, no waiting, nothing to press.
<5s
05 — Private by design
Nothing leaves your computer.
- No accounts, no API keys. You never sign in to anything.
- No cloud. Your files are never uploaded — the thinking happens on your own machine.
- Works offline. Even on a plane. There is no network at question time.
- Only your computer can talk to it. It listens on 127.0.0.1 — a door that only opens from the inside.
the internet— nothing crosses this line
06 — Pictures too
Find a picture by describing what's in it.
No tags, no file names needed. It looks at what the image shows — try one:
[ architecture diagram ]match · 0.33
[ whiteboard photo ]match · 0.33
[ dog on the sofa ]match · 0.33
[ harbor at sunset ]match · 0.33
It isn't reading text inside the image — it's recognising what the picture depicts.
07 — Receipts
Every question gets a receipt.
Semlith writes down what your AI looked up. Each line is chained to the one before it — so if anyone changes a line, the chain visibly breaks. Click a line to try.
✓ All 6 lines check out. Each one carries a fingerprint of the line before it.
08 — In numbers
Quick, private, and fits in wherever you work.
46
languages it reads — code, markup, config
27
AI assistants it plugs into
13
document types — PDF, Word, slides, sheets, notebooks…
16 ms
to answer a search, warm
<5 s
from save to searchable
0 bytes
of your files sent to the cloud. Ever.
1
line to install
Index once.Ask anything.Keep it all at home.
SEMLITH · RECEIPT
your visit, written down
visitoryou
readthis page
time here0 s
sections seen1 / 10
bytes sent anywhere0
chain-------- ✓
thanks for reading
One line to get it
No account. No keys. Nothing to sign up for.
$ curl -fsSL https://raw.githubusercontent.com/semlith/semlith/main/install.sh | shmacOS (Apple silicon) · Linux · Windows — then it sets itself up for the AI tools you use.
github.com/semlith/semlith open source · Apache-2.0built in Rustno network at question time