01Remember

The gap

Your AI forgets you when the chat ends

Every new chat starts from zero, so you paste the same context again. What you told one chat is gone for the next — unless it lives somewhere every session can reach.

The knowledge is already yours. It just isn’t where your agent can find it.

Lyt closes that gap without moving your notes anywhere. They stay files you own; your agent gains a way to reach them.

One first win

Remember it once. Get it back later.

Tell your agent to remember something once, and it can bring back the original note later. Here is the whole loop, in three moments and in your own sentences.

One written document strobed along a single path across one continuous surface: entering shelter at the left, at rest in the middle, and emerging into the open at the right.

You say: save this

“Save this decision — we're going with Apex Print as our printer supplier, because they can hit our deadlines.” Your agent writes one plain markdown file into your own notes: a title, a date, and a line saying why it was worth keeping.

Weeks pass

You forget the reasoning. That is normal, and it is the exact failure this is built for. The note sits in your own folder the whole time, in a format you can open in any editor, with or without Lyt.

You ask, later

“What did we decide about the printer supplier, and why?” Your agent searches what you wrote and comes back with the decision — and names the note it came from, so you can open the original yourself.

Lyt itself sends nothing anywhere in that loop. Your agent starts from what you decided: it works from your decisions, notes, and documentation — not from what it can guess. See the loop in full.

02Understand

What Lyt is

Your knowledge, in files you own

Your notes stay ordinary markdown files in ordinary folders. Lyt gives your agent a reliable way to find them and use them.

Vault

A folder of markdown files with a name of its own — usually one per subject: work notes, research, a team handbook.

Mesh

How you group related vaults, and how you declare which ones are connected to which.

Pod

Everything this machine can reach — your own vaults, the ones shared with you, and the ones you follow — as one place to search.

Already keep notes in Obsidian? That vault is already plain markdown — Lyt adopts it as it stands, nothing moves. Adopt what you have shows how.

For the technically minded: a vault’s identity survives rename and move; meshes declare relationships; the pod is the complete set you can reach, shown on each machine as one searchable view. See how the model fits together.

You never operate any of this by hand: you ask in plain language and your agent runs the right operation. Setup teaches it Lyt ahead of time — a manual and a set of skills written into its own instructions — connect an agent shows what it learns.

03Own

It stays yours

Your memory lives in files you control

Your notes stay files you can open, move, and keep — nothing moves into a hosted database.

A slight ruled index stands in front of a deep stack of written pages, plainly derived from them and plainly the lesser object.

Markdown is the source

Your notes remain ordinary Markdown files in ordinary Git repositories. Lyt builds disposable local indexes around them.

One compact drive stands at rest on a surface, a single small green indicator lit low on its near face.

Local-first

Capture, search, and indexing work locally. Matching by meaning runs through a small model on your own computer, downloaded once if you choose.

The Lyt vault form standing shut and sealed on a plain surface, closed and private.

Nothing reports home

No hosted knowledge service, no passive usage stream. Network activity follows an operation you choose.

A written page carrying a blind-embossed seal, its concentric rings and ridges raised in the paper itself and catching the light.

Apache-2.0

The seven-package toolchain is open source. Read it, fork it, or keep working with the files without Lyt.

Under the hood, Lyt adds stable identity and local indexes around your files — it never replaces them.

04Find

Finding things again

Half-remember it. Your agent finds the matching notes.

Every note that comes back names where it came from — the vault, the file, your own words.

A wide field of dim, identical note tiles laid out in rows. One tile has risen clear of the field and is lit bright, hovering above the green-glowing gap it came from.

What comes back is what you wrote

Lyt retrieves and ranks your own notes. Notes your pod has already organised around a topic come first; full-text matching backfills the rest. The note you get back is the note you wrote.

And nothing leaves your machine

Indexing and searching both run locally — your notes are not sent anywhere to be searched. Matching by meaning rather than wording is available once you choose to download a small model, which also runs on your own computer.

Search & indexing describes the mechanism.

Start with the model

Two pages before anything is installed

What Lyt is, and what people do with it — written for a first read, not for an operator.

A precision linear rail and vacuum plate lifting a written document from a glass-shelved vault, a gathering of pale light resting at the working head where the action happens.

How Lyt works

Vault, mesh, pod — the model in plain language, and what your agent does with each one.

Read how it works

A wide facility floor seen from above: vaults floating in ranks over their pads, holographic agents standing at stations, transport lines carrying documents between them.

What people build on Lyt

A personal pod your agents read, a team sharing one source, published knowledge others subscribe to, and the same pod on every machine.

See the use cases

05Connect

Then a team

Share it with the people you choose

Nothing is shared on its own. You grant each person access, by name, and your team works from one shared source.

In near darkness, a heavy dark ring encloses a glass case. One narrow door in the ring stands open and throws a single beam of light outward to two small blocks waiting in it; the rest of the ring stays shut.

Your team works from one source

One versioned knowledge base, explicit boundaries, and every teammate's agents searching the same material.

Access is granted, never inferred

You grant each person read or write access where the vault is stored online. Grouping vaults together never grants anyone access on its own.

Someone new starts from the material

A new teammate reaches the same decisions and documentation their colleagues already work from, instead of being told them a second time.

Share a vault walks through granting access. A team knowledge base shows what it looks like in use.

The same motion goes outward: when someone else’s knowledge is already public you can follow it, and your copy keeps naming where it came from. Everything Lyt creates starts private — making a repository readable by anyone is a step you take yourself, where it lives, and no Lyt command changes visibility. Sharing sets out owned, shared, public, and subscribed precisely.

Get Lyt

Ask your agent to set up Lyt

Paste this request into any coding agent you already have. It spells out every step, and everything it creates stays local until you choose to share.

Set up Lyt on this machine. Lyt is the npm package @younndai/lyt. It requires Node.js 20.9 or newer, and npm.

After setup, the quickstart keeps the path small: create or adopt a vault, capture a note, find it again.

Follow the quickstart

Under the hood. Your Markdown stays the source — declarations record identity and topology, and local indexes are derived around your files, never in place of them. A typed tool surface is available for clients that prefer one: 17 tools running as a local child process, with no hosted endpoint and no port to expose. The 7 packages are released together, so they always match.

Connect an agent has the configuration; the package architecture lists what ships.