Skip to content

Chrome extension · 15 languages · Nothing auto-sends

Select text.Act in one click.In their language.

Typeglass lives in Chrome. Highlight a message, image, or field — translate, digitize, reply, fix grammar, or write from your knowledge base. It proposes; you decide. Nothing is sent until you say so.

Get startedSee how it works
  • 500 free AI runs a day
  • No API key
  • No card

The extension is not on the Chrome Web Store yet — sign up and the same writing and vision tools run in your browser at /agent today.

languages supported
15
capabilities in one popover
12
messages ever auto-sent
0

Hello in 15 languages: English, French, German, Spanish, Italian, Portuguese, Dutch, Polish, Russian, Japanese, Simplified Chinese, Traditional Chinese, Korean, Arabic, Hindi.

How it works

Select. Act. Confirm.

Built for Chrome first. The website is where you sign in, tune preferences, grow a knowledge base, and steer the roadmap — not where you do the daily work.

  1. 01

    Select

    Highlight text and press ⌘⇧K (Ctrl+Shift+K), right-click an image, or snip a region on any page — Gmail, Slack, WhatsApp Web, Teams, Discord, or a bare textarea.

  2. 02

    Act

    One popover: translate, digitize, reply, grammar, write, or pull from your knowledge base — without leaving the tab.

  3. 03

    Confirm

    Every result is a preview. Inserting it is your click or keystroke. Typeglass has no ability to send on your behalf.

In the popover

Translate, reply, fix, write — in any of 15 languages.

Reply in their language

Detects the language of the message you are answering and drafts your reply in it — or bilingually, with the same reply in your own language underneath, so you know what you are sending.

Translate what you select

15 languages end to end — read a message in theirs, or write your draft in yours and translate before you insert. History stays on your device.

Match the tone

Professional, Friendly, Concise, Detailed — or type your own instruction. “Apologetic but firm.” “Ask for a meeting next week.”

Fix grammar, keep your voice

Corrects spelling, punctuation and grammar without rewriting your wording into someone else's style. Works in all 15 languages.

Write from your knowledge base

Save the phrases, policies and product facts you reuse. Typeglass drafts with your voice and context — not a generic model guess.

Hand over a whole task

reasons in steps

Describe a goal — “summarize this thread, then draft a friendly reply in French” — and the agent plans, calls its tools across up to 8 steps, and shows each one as it works.

Digitize

Anything you can see becomes material you can edit.

A screenshot, a whiteboard photo, a chart, a table trapped in an image — right-click it or snip a region, and it comes back as an editable diagram, working text, or real data. In the same popover as the writing tools, pre-filled and ready to go.

Pull text out of any image

Pick an image file, drag one in, right-click an image, or snip a region of the live page — including PDFs and local files. The text lands in the popover, ready to work on.

Turn a photo into a diagram

reasons in steps

Right-click a whiteboard sketch or flowchart. A vision model picks the diagram type that fits and rebuilds it as an editable diagram — then open it in draw.io, Excalidraw or Mermaid Live.

Digest a long page

reasons in steps

Explain or summarize the whole page, not just the top. Long pages are chunked, condensed in parallel, then synthesized — with an optional focus.

Capture

Record, transcribe, speak and picture it.

Not everything starts as text. A screen recording to explain a bug, a voice note to answer, a reply you would rather say than type, an image for the post — the same extension handles them, in the same popover.

Record your screen, the Loom way

A screen, a window, a tab, or just a region you draw — with your microphone, the tab's audio, and a camera bubble you can drag. Pause, trim, export a GIF, download or copy it. Recordings stay on your device, need no account, and never use a run.

Transcribe an audio file

Pick a voice note or a meeting clip and get the words back as text you can translate, summarize or reply from — in the same popover.

Speak instead of typing

The microphone in the popover turns what you say into the text box, so a reply can start as a sentence spoken aloud.

Generate an image

Describe it and pick a style — Photorealistic, Line art, Flat illustration or Watercolour. Then copy it, or download it as a PNG.

Remember

It learns how you write — and where things go.

Memory, writing skills and your knowledge base shape every draft, so the tenth reply sounds more like you than the first. Smart Bookmarks keeps what you save findable.

Memory that carries between tasks

Short facts the assistant should keep — “I prefer British English”, “My team is Orbit” — replayed with every agent run. Saved on your device; signed in, they follow your account.

Writing skills that sound like you

Style rules woven into every reply, translation, rewrite and agent task, so responses read like you wrote them. Two recommended rules are on by default; add your own.

Smart Bookmarks

Press ⌘D (Ctrl+D) and the page is filed into a folder you already have — never one it invents. It learns from your corrections, and it runs on your device by default.

Where it appears

In Chrome, on the page — from Gmail to Teams to Discord.

The Chrome extension is where Typeglass lives day to day. Built-in adapters read what you are replying to on the first eight surfaces; a generic adapter covers everywhere else you can place a cursor. Signed in, the website's /agent runs the same writing and vision jobs in a tab — preferences, knowledge base, and roadmap live here too.

  • Gmail
  • OutlookAdd a message
  • LinkedInAdd a comment…
  • X (Twitter)Post your reply
  • SlackMessage #design
  • WhatsApp Webاكتب رسالة
  • TeamsType a message
  • DiscordMessage #general
  • GitHubLeave a comment
  • NotionType “/” for commands

And when a site has no adapter at all? Any bare <textarea> works — if you can place a cursor in it, Typeglass can write in it.

How it behaves

It proposes. You decide.

Six guarantees about who is in control — designed into the product, written down here so you can hold us to them.

0

messages ever auto-sent — an absence of capability, not a setting.

  • Every result is a preview

    Inserting a result is always your own click or keystroke, and Typeglass has no ability to send a message on your behalf — not a setting you can flip, an absence of capability.

  • History stays on your device

    Your drafts, history and screen recordings live in your browser's local storage and are never uploaded. Signed in, your settings and saved memories follow your account so a second browser starts where the first left off — and one button deletes that copy without touching this device's.

  • Message content is not logged

    Our servers record that a request happened and how large it was, never what it said. Images are read for that one request and not retained. The one identifier in that line is the key the request is metered against — your account id when you are signed in, one derived from your IP address when you are not.

  • No API key to manage

    Model access is included in every plan, Free included. Nothing to create, fund, rotate, or discover has expired in the middle of a customer conversation.

  • On-device AI when you want it

    Where Chrome can run its built-in models, Translate, Summarize and Extract text can run on your computer instead. Switch on private mode and they never reach our servers — if a model is not ready, they stop rather than send anything.

  • Sensitive sites are off by default

    Major banking, healthcare and password-manager sites are blocked out of the box, and you can turn Typeglass off on any other site.

Plans

Start free. Grow by the day.

Every AI action is one run, and every plan is a number of runs a day. Recording, bookmarks, history and settings never use one. The count resets at midnight UTC.

  • Free

    500runs a day

    Every feature, on your own account. No card.

  • Pro

    5,000runs a day

    10× the Free allowance, for writing all day.

  • Team

    20,000runs a day

    40× the Free allowance. Allowances belong to one person each and are not pooled.

Compare plans

Beyond the extension

In a tab, or in your editor.

  • In a browser tab

    The same tools at /agent

    Signed in, the website runs the same writing and vision jobs in a tab — threads, attachments, dictation and a capture bar — for a computer where the extension is not installed.

    Open the web agent

  • In your editor

    A self-hosted MCP server

    Separate from the extension and its plans: a developer who runs their own Typeglass server can also run it as an MCP server exposing 13 tools — 7 writing, 6 vision and data — for VS Code, Cursor and other MCP hosts. It runs on its owner's own model key; the extension never asks you for one.

Questions

Before you get started.

The questions people ask first, answered before anything is asked of you. Missing one? The address in the footer reaches a human.

Is this a website or a Chrome extension?

The product is the Chrome extension — select text or an image on any page and act in one click. Signed in, this website also runs the same writing and vision jobs at /agent, and holds preferences, your knowledge base, and feature requests.

When can I actually install it?

The extension is not on the Chrome Web Store yet, and there is no date we would trust ourselves to promise. You can start today: create a free account and the same writing and vision jobs run on this website at /agent. When the listing goes live, “Get started” at the top of this page becomes “Add to Chrome”.

Does it cost anything to start?

No. The Free plan is 500 AI runs a day with model access included, and requesting a feature is free too. Larger daily allowances are listed on the pricing page, and nothing is charged unless you choose one.

What happens when I reach the daily limit?

AI actions are refused until the reset at midnight UTC, and the message says when in your own time. Everything that does not use AI — recording, bookmarks, history, settings — keeps working. There are no overage charges.

Which languages does it support?

English, French, German, Spanish, Italian, Portuguese, Dutch, Polish, Russian, Japanese, Simplified Chinese, Traditional Chinese, Korean, Arabic, Hindi — 15 in all, for reading, replying, translating and fixing grammar.

Do I need an OpenAI key?

No. Model access is included — there is no key to create, fund or rotate. That is most of what you would be paying for over running something yourself. Only a self-hosted Typeglass server (the developer option above) runs on its operator's own key.

Does it work on sites you have not heard of?

Yes. Built-in adapters read the conversation on Gmail, Outlook, LinkedIn, Slack, WhatsApp Web, Teams and Discord, and the post you are replying to on X; a generic adapter handles everything else. If you can type in it, Typeglass works there.

Ready when you are

Write in their language from the next tab you open.

Typeglass proposes; you decide. Nothing is sent until you say so, and your history stays on your device.

Get startedSee how it works