Launching November 18

A personal AI
that does real work.

Talk on a voice call, hand off background work, get meeting notes with speaker labels, and reach it from your apps and channels. It lives at the top of your screen.

macOS · Windows · Linux

TameSonnet 4.6

Tame

Anthropic · claude-sonnet-4-6
Linear open an issue for the flaky checkout test
Tame
Searched 42 files with a read-only subagent
Created LIN-142 in Linear

Done. LIN-142 tracks the flaky retry in checkout, with the failing run linked and assigned to you.

+ GoalWhat should I say?
GIFAsk me or mention @companion…

Meetings

41:12· 3 people
ComputerMicrophone
TranscriptAnswersNotesSummary

QuestionSo we ship the new onboarding on Thursday?Alex · asked

YouYes, if QA signs off by Wednesday noon.

SamI'll own the release notes and the changelog.

What should I say? Ctrl 1Follow-up questions Ctrl 2Recap Ctrl 3Auto answer Ctrl 0
Ask about this meeting…Ctrl ↵

Apps

Connect Notion, Linear, GitHub and any remote MCP server.

Recommended for you
GitHubRepos, issues, pull requests and Actions.
LinearIssues, projects and cycles.
NotionPages, databases and search.
SentryErrors, issues and releases.
More apps
1PasswordVault and item metadata, read-only.
CloudflareCloudflare docs, Workers and Pages guidance.
FigmaFiles, components and comments.
PostHogInsights, funnels and feature flags.
StripePayments, customers and invoices.
VercelProjects, deployments and logs.

Channels

Connect a channel

Talk to your companions from Telegram, Slack, Discord and more. Each chat has its own history.

Pairing requests
Maya ChenStudio Telegram · Code 7Q4K2MDismissGuestIt's me
Connected chats
Slack · DesignSlack · Answers as TameOffManage
Studio TelegramTelegram · Answers as TameManage
Approved senders
Ana SouzaStudio Telegram · Answered with guest permissions.It's meGuestBlocked

Whoever writes for the first time gets a code for you to approve here.

Interactive preview · sample data

Hand it the work.
It handles the rest.

Voice calls

Start a call.
Just talk.

Start a local voice call from the notch and talk to your active companion. Each companion keeps its own conversations, and it remembers your facts and preferences.

  • Interrupt, mute or use push-to-talk
  • Bring in a meeting, an app or the vault with an @mention
  • Council: up to six models advise the one that answers
Background tasks Routines on Windows & Linux

Hand it a task.
Get back to yours.

Send a background task and keep working. It tells you it's on it, you can cancel any time, and the result posts back to the chat when it's done.

  • Says it's on it, and stops when you say stop
  • Subagents for the heavy digging; Stop cancels every child
  • Routines on a schedule, a webhook or a poll, and skills it learns once you approve them
Meetings Windows & Linux

Meeting notes
with speaker labels.

Tame listens on your computer. It does not join the call. It captures your mic and computer audio, labels speakers, and leaves you with notes and follow-ups. Opt in and it gives you live help on what to say next.

  • Decisions and action items, each linked to the lines that back it
  • Replay at 0.75–2×; export to Markdown, text or JSON
  • Google Meet names on speaker labels via the Chrome or Edge extension
Apps & channels Channels on Windows & Linux

In your apps.
On your phone.

Your companion opens the issue in Linear, publishes in Notion or reviews the PR. Message it from your phone on 40+ channels like Telegram, WhatsApp, Slack, Signal, iMessage and email.

  • 286 verified apps, plus your own MCP servers on Windows and Linux
  • Telegram, WhatsApp, Slack, Teams, Signal, iMessage, email and more
  • Same companion and memory; each chat is its own conversation
286apps in the catalog,
a click or a token each
80+AI providers behind
one picker
40+messaging channels,
WhatsApp and iMessage included
20interface languages,
Brazilian Portuguese included

286 apps.
One connection each.

Each app is checked against its live server. Most connect with one browser sign-in; a few take a token. On Windows and Linux you can add any MCP server too. A few of them:

  • Notion
  • Linear
  • GitHub
  • Sentry
  • Figma
  • Atlassian
  • Asana
  • GitLab
  • Stripe
  • Neon
  • Supabase
  • PostHog
  • Vercel
  • Cloudflare
  • 1Password

Expand the island
into a studio.

One conversation per companion, answers that render as real cards, charts, tables and forms, and a private Linux computer when the job needs one. This sample runs itself.

Which team spent the most time in meetings this week?

Here's the split from this week's meeting notes. The numbers are fictional; the card is real.

Meetings this week

Interactive sample · fictional data
Engineering12 hHours in meetings · this week
Hours in meetings · sample
Compare the teams with a table.
TeamHoursMeetings
Design6 h4
Engineering12 h7
Sales18 h11

Tell me what to change, and I'll add it to your launch checklist.

Engineering · cut one status meeting.

On it. I'll draft the change in Notion. Meanwhile I'm running the checkout test on my own computer while you read the summary — one moment.

Checkout is green: 42 passed, one flaky retry. Want me to open the pull request?

Scripted product sample with fictional data. Cards, tables and forms are rendered from a fenced openui-lang block — a card can ask, never approve. The bot computer is a real, isolated Linux desktop on Windows and Linux, in your InstaBox account or a local container.

Questions,
answered.

What does Tamekoshi actually do?

It's a personal AI you hand real work to. It lives at the top of your screen. Talk on a voice call, send background tasks, get meeting notes with speaker labels (it listens on your computer and does not join the call), work inside apps like Notion, Linear and GitHub, and reach it from Telegram, Slack or iMessage. Watching your coding sessions is part of the toolbox, not the whole product.

Can I talk to it on a voice call?

Yes. Start a local voice call from the notch and talk to your active companion. On Windows and Linux there are 10 voices in 5 languages; Mac uses its system voices. You can interrupt, mute or use push-to-talk.

Can it work in the background?

Yes. Hand it a background task: it tells you it's on it, you can cancel, and the result posts back to the chat. Subagents work the same way. Stop cancels every child.

What is the Council?

A way to mix models. Pick Council as a companion's model: up to six models weigh in, and one speaker answers. On Windows and Linux.

Which coding agents are supported?

Claude Code everywhere. Cursor and Codex get their own pills on Windows and Linux; Gemini CLI and Antigravity on Mac. Any other agent can join by tagging its hook payload. On Windows and Linux, your companion can also hand a whole coding task to Claude Code, Codex or OpenCode in its own git worktree.

Which AI models can I use?

80+ providers behind one picker: Anthropic, OpenAI, Google, OpenRouter, Ollama and more, plus OAuth sign-ins like GitHub Copilot and xAI. Bring your own key or sign in; each companion can run a different model. On Mac today: Anthropic, Google and OpenAI.

Where can I message it from?

On Windows and Linux, 40+ channels, including Telegram, WhatsApp, Discord, Slack, Teams, Matrix, Signal, iMessage, email, SMS and IRC. Same companion, same memory. Mac does not ship channels yet.

What's in the apps catalog?

286 apps in nine groups, like Notion, Linear, GitHub, Sentry, Figma and Stripe. Most connect with a browser sign-in; GitHub, Figma and Stripe take a token; Cloudflare needs no login; 1Password stays read-only. On Windows and Linux you can add any MCP server.

Does it ask before it acts?

Yes. Reads run, writes ask, and money or destructive actions always ask. Auto-review and approval rules handle the routine calls, and your companion can tune those rules with you from chat. Coding-agent hooks never block: if Tamekoshi isn't running, they exit instantly.

Can it work on a schedule?

Yes. Ask for a routine in chat, like a morning brief or a weekly digest, and it runs on its own. Background work reports back when it's done, and after a restart it picks up a recent task once. Routines run on Windows and Linux.

What's the companion's computer?

A private Linux desktop (and its own browser) that belongs to one companion, on Windows and Linux: a machine in your InstaBox account or a local Docker or Podman container. It can run builds, browse and report back. You can take over or put it to sleep whenever you like. Mac does not include the companion computer or browser yet.

What's different on Mac?

Mac has voice calls, chat with Anthropic, Google and OpenAI, the vault (Touch ID), apps, skills and coding-agent monitoring. It does not yet include meetings, channels, routines, the Council, rooms, or the companion's computer and browser. Those ship on Windows and Linux today.

My Mac doesn't have a notch. Can I still use it?

Absolutely. On an iMac, a Mac mini or a MacBook with the lid closed, it sits in a small bar at the top of the screen. Same on Windows and Linux.

Which platforms does it run on?

Mac, Windows and Linux. On a Mac (macOS 15 or later) it lives in the notch, or in a slim bar if there's no notch. It runs on Windows 10 and 11, and on Linux with Wayland or X11. Some features are still Windows and Linux only; see the Mac question above.

How long does setup take?

A couple of minutes. Download and launch, pick Tame's look and name, add an AI provider, and connect your coding agents. The rest is already sensible.

When can I try it?

We launch on November 18. Join the waitlist below and we'll email you when your build is ready.

Your companion is
almost home.

Launching November 18. Be first to know when your build is ready.

By joining, you agree to receive early access and launch emails. See our Privacy Policy.