Skip to main content
On-device AI — private & free forever

Keep it Glassy

Your second brain. Uncompromised.

The workspace for people who think in 47 tabs, 12 notebooks, and a phone full of unlabeled voice notes.

Local-first Files on your disk. Always. No tracking. No ads. No training. Markdown-native. Export anytime.

See Glassy in 100 seconds

1:40 Product tour
Download ↗

Sound familiar?

You're not disorganized. You just have more ideas than a single tool can hold.

Forty-seven tabs. Three of them important.

Save the page, not the tab. GlassyKeep tucks links into searchable collections — and the Companion extension does it in one keystroke.

A voice memo at 2 a.m. you'll never find.

Voice Studio transcribes locally with Whisper — your audio never leaves the browser. Search the transcript, not the silence.

Eight hundred saved links. Three of them read.

Read-later that actually surfaces. GlassyKeep tags, summarizes, and groups by meaning — so the good ones come back to you.

Notes in five apps. None of them home.

One page for the daily note, the long doc, the quick capture, and the napkin idea. Your Second Brain indexes everything so you can query it from Claude, Cursor, Windsurf, OpenCode, Hermes, Openclaw, Pi, or any MCP agent. File-based. Portable. Yours next decade too.

Glassy isn't another app on the pile. It's the page you actually open.

One workspace. The whole job.

Capture, think, surface — in one calm page. Stop app-hopping. Start finishing.

Free Forever

AI that stays on your device.

Grammar, tags, summaries — free, local, forever

Qwen 2.5 and Whisper run inside your browser via WebGPU. Grammar fixes, tag suggestions, voice transcription — no API keys, no credits, nothing uploaded. Cloud AI is opt-in for the harder work, and every free account gets local AI for life. Your Second Brain knowledge base is queryable by Claude, Cursor, Windsurf, OpenCode, Hermes, Openclaw, Pi, and any MCP-compatible AI agent via a real MCP server.

AI writing assistant inline in a Glassy document, showing expand, simplify, and grammar-fix actions
Pro+

Talk it out. Read it back.

Local Whisper · your audio never leaves the browser

Record a memo, get a clean transcript, let AI lift the action items. The whole pipeline runs on your machine — not a single byte of audio is uploaded.

Voice Studio recording interface with waveform visualization and live transcript
⌘K

Find any thought in two keys.

Text + meaning, in one search

Notes, documents, voice transcripts, bookmarks — ⌘K searches them all at once. Match by exact words or by what you actually meant. If you remember the shape of the idea, Glassy finds the file.

Universal Search ⌘K overlay showing mixed results across notes, bookmarks, and documents

Documents that read like you write.

Slash commands, callouts, embeds, doc icons

Press / for headings, callouts, code, and rich embeds. Give each doc its own icon so you spot it instantly. YouTube, Spotify, any media link drops in as a native widget — no more "open in new tab."

Rich TipTap editor with slash command menu open, showing heading, callout, and embed options
Pro

GlassyKeep + Companion.

A read-later that actually surfaces

One-keystroke save from the Companion extension. Screenshot any page and embed the image directly in your note. AI auto-tags and summarizes. Smart Collections group by meaning. Search the eight hundred bookmarks you forgot you had — the good ones come back.

GlassyKeep masonry grid showing bookmarked articles, videos, and research with AI-generated tags
Deep Integration

Your Obsidian vault, in any browser.

Browse, search, push & pull — the foundation for AI agent memory

Connect your vault and Glassy respects every folder, every wikilink, every frontmatter field. Read and search from any browser. Your files never leave your disk. Now live: query your vault through the Glassy MCP server from Claude, Cursor, Windsurf, OpenCode, Hermes, Openclaw, Pi, or any MCP agent — no separate install.

Obsidian vault browser showing folder tree, markdown files, and wikilink navigation

Write together. Cursors and all.

Real-time multi-cursor collaboration

Invite a friend or a teammate. See cursors fly in sync. No ceremony — just a link and you’re both in the doc.

Real-time collaborative editing with multiple cursors and live presence indicators

RSS, without the doom-scroll.

Image-forward feed reader, built in

Curated feeds, custom URLs, or any Public Window. Read on your terms — scroll what you want, when you want, no algorithm telling you to stay.

RSS feed reader with image-forward article cards and clean reading layout

Built in the open

Trust the trail,
not the pitch.

Glassy is still young. So we would rather show what shipped, what changed, and what the product refuses to do than decorate the page with borrowed certainty.

Glassy is the first app that doesn't punish me for thinking sideways. The Second Brain is the only one that remembers what I meant, not what I typed.

Founding beta tester · Wendy W. · Day 142

Changelog first

Every meaningful change leaves a trail.

Pricing, local AI, privacy, storage, and extension work are written down as they ship, so the story does not drift away from the product.

Open the changelog

MCP Server

A real knowledge layer, not vaporware.

Fifteen tools, three prompts, six resources — built on the official MCP SDK v2. Query your corpus from Claude, Cursor, Windsurf, OpenCode, Hermes, Openclaw, Pi, or any MCP-compatible client. DNS-rebinding protection. Tier rate limits. It ships, it works, it is documented.

See the Second Brain

No growth tricks

No streaks. No red badge theater.

The product is designed for return, not retention hacks: capture first, organize later, and keep the interface quiet by default.

Read the design principles
ND-affirming by design

Quieter by design. Kinder by default.

ADHD, autistic, AuDHD — every brain that does its best work in bursts deserves a workspace that doesn’t punish the burst. Glassy’s defaults come from real research on executive function and sensory processing, not productivity Twitter.

Read the design principles

Capture first, decide later.

One keystroke opens a blank page. Voice, sketch, text, clipboard — into the workspace in under three seconds. No empty folders to fill first.

Scannable by sight.

Color-coded cards, emoji icons, distinct shapes, and curated themes. Read the layout, not just the words.

Cognitive scaffolding, not surveillance.

On-device AI cleans up brain-dumps, expands bullets, and turns memos into tasks — without sending your day to a server.

No shame mechanics.

No streaks weaponized against you. No angry red overdue badges. Reduced-motion respected. Archive over delete. Your pace is the right pace.

Your data never leaves your browser

Your AI stays on your machine.

Most AI tools route your thoughts through someone else’s server. Glassy runs real models inside your browser via WebGPU. No cloud required. No credits consumed. Nothing leaves your machine unless you ask.

Running today

Qwen 2.5 (0.5B / 1.5B / 3B)

Instruct-tuned local LLMs. Summarize, rewrite, tag — entirely in your browser tab.

Whisper Base + Small

On-device speech-to-text. Voice memos arrive as cleanly written notes.

MXBai + MiniLM (384-dim)

Local embedding models power semantic search and tag clustering without a server.

How it actually works

WebGPU acceleration

Models run on your GPU, not someone else's data center.

Two runtimes

transformers.js + WebLLM, picked per task for speed.

Smart routing

Glassy picks the smallest local model that handles the job. When a query needs a larger context window or a harder task, it routes to your configured cloud provider — OpenAI, Anthropic, or any OpenAI-compatible endpoint. Cloud is opt-in; you stay in control.

Download once

Cached in IndexedDB. Works offline forever after.

Chrome built-in AI (Tier 0)

On Chrome 138+ with the Prompt API enabled, Glassy uses Gemini Nano — zero download.

This is genuinely new territory. We're excited to be here.

Local AI in the browser is bleeding-edge. Models are getting smaller and more capable every month. Browser GPU APIs are still maturing. Here's exactly what's on our radar.

Gemma 4 E2B / E4B

Tracking WebLLM prebuilts

Newer Google open weights designed for on-device use. Native audio input + structured JSON. Transformers.js already supports the architecture — waiting on WebLLM WASM prebuilts to ship (no ETA as of June 2026).

EmbeddingGemma

Drop-in upgrade path

Higher-quality on-device semantic embeddings from Google. Direct upgrade target for the MXBai/MiniLM tier — same 384-dim local pipeline, better recall.

Moonshine ASR

Evaluating for Voice Studio

More efficient than Whisper for live transcription. Smaller footprint, designed for streaming audio. Potential drop-in for the Voice Memo pipeline.

The real questions.

The objections we hear most often — answered honestly, in plain English.

What is the Second Brain?
Your bookmarks, notes, documents, voice transcripts, and Obsidian vault files — automatically indexed into a searchable knowledge base. Query it from Claude, Cursor, Windsurf, OpenCode, Hermes, Openclaw, Pi, or any MCP-compatible AI agent. Your corpus stays on your server; only the specific query leaves your machine.
How is this different from Obsidian?
Glassy is not trying to erase your vault. It gives Markdown-shaped work a calmer web surface, local AI, capture tools, and sharing when you want them.
Is local AI actually local?
Supported AI actions run on your device when your browser and hardware allow it. Cloud AI is explicit, optional, and reserved for work that needs more than the local model can handle.
What leaves my device?
Your account data syncs when you use hosted Glassy. Local AI prompts stay on-device for local actions; cloud AI requests only leave when you choose that path. When a query routes to cloud, only the relevant captures for that query are sent — not your whole library.
What happens if I want to leave?
Export to Markdown and keep your files. Glassy should make your work more useful, not hold it hostage.
Can I self-host?
Self-hosting is part of the Glassy path. The app ships as a Docker-based stack and we document the process honestly — every step, every tradeoff, nothing glossed over.
Does the Companion extension work in Firefox?
Yes. Glassy Companion supports Firefox 121+. All capture modes — Quick Save, highlights, element pick, and screenshots — work in Firefox. The Side Panel feature (Ctrl+Shift+P) is Chrome/Edge-only; Firefox users use the popup instead.
Where are my screenshots stored?
Screenshots are stored on your Glassy server — `app.glassy.fyi` for cloud accounts, or your own server for self-hosted instances. They live at uploads/captures/ on the server and are embedded as standard Markdown image links in the capture note. Nothing goes to a third-party image host.
Does Glassy work offline?
Yes, and it keeps your captures safe. If you save something while your Glassy server is unreachable, the Companion extension queues it locally and syncs automatically when the connection returns. You will see a badge on the toolbar icon showing the number of items queued. Nothing is dropped.
Is this an ADHD app or an autism app?
Both, and neither. It is a workspace that respects nonlinear brains without making accommodations feel like a separate mode.
What's the catch with the free tier?
There is not one. Free stays useful; Pro funds the heavier workspace pieces and cloud costs.
What is Clear?
Clear is the all-inclusive tier for people who want every current and future theme pack plus the highest-touch community access. Most people should start with Free or Pro.
Can AI assistants ingest your docs?
Yes. See /llms.txt and /llms-full.txt. We welcome crawlers that cite sources and respect publisher intent.

One free tier. Two paid tiers. No rush.

Start free. Move to Pro or Clear the moment the full workspace is worth it to you.

Checkout can happen first. Create or sign in with the same email after purchase to claim your workspace.

Free

Everything you need to think clearly.

$0 forever
Storage 10 MB
  • Notes, documents, and daily pages
  • On-device AI — free for every account, forever
  • Second Brain / MCP server — 120 queries / hr
  • Real-time sync across browsers
  • Universal search (⌘K) by text and meaning
Start for Free
Most popular

Pro

The whole workspace. One price.

$100 / year

≈ $8.33/mo · 2 months free

or $17/ month

Storage 200 MB
  • Everything in Free
  • GlassyKeep + GlassyCalc sidecars included
  • Voice Studio — local Whisper transcription
  • Second Brain / MCP server — 1,200 queries / hr
  • Cloud AI included · top-ups available
Choose Pro

Cancel anytime. Data retained 30 days after cancellation.

Membership

Clear

A private community inside Glassy. Early access to every new tool.

$300 / year

≈ $25/mo · the whole community

or $80/ month

Storage 1 GB
  • Everything in Pro
  • Every current and future theme pack, included
  • Private member community + feed, inside the app
  • The unrestricted single-user Docker build — yours today
  • Every new feature and AI tool, first
Become a Founding Member

From the blog

Notes on building quiet tools