Sign in
✦ Now in its own window on Mac and Windows

Every AI.
One app. One memory.

Ask Claude, ChatGPT, Gemini and hundreds more at once — side by side. They all share one private, encrypted memory that lives on your own computers, so you never explain yourself twice — and if one AI service is down, another carries on the same conversation. Start in your browser in a minute — nothing to install.

Start chatting free
  • 🛡️ Never read, shared or used for AI training
  • 🔒 End-to-end encrypted
  • 🌐 Browser, phone, Mac and Windows
  • 💸 Free AI, or the subscriptions you already have

Your data stays yours: your chats are encrypted, KumbayAI never shares them with third parties and never uses them to train AI models. The fine print.

  • ✦ Ask All AgentsOne question to several AIs at once, with one combined answer and every full answer underneath.
  • ◫ Split ChatsUp to four chats side by side, from any mix of agents. Send one message to all of them.
  • 🧠 Memory OSEvery agent shares one private, encrypted memory of you — switch AI, keep your context.
  • ∞ Free AIChat free: all your free keys as one agent that balances their limits.
  • ⇄ Stand-in agentsAI service down? Carry on with another agent, then switch back when it returns.
  • ▣ Its own windowOn Mac and Windows, KumbayAI opens in a window of its own, from the menu bar or tray.

On by default

Memory OS — an AI that remembers you, whichever AI you use

Today every AI starts from zero. With KumbayAI’s Memory OS, your agents share one private memory of you: your preferences, your projects, the decisions you already made. Teach it once to Claude and Gemini knows it too.

Remembers for you

Once a chat goes quiet, KumbayAI quietly turns it into short memories — preferences, facts, decisions — and chat summaries you can search by meaning, not just by words (Hebrew word forms too). Chats you import from other AI apps are added to memory automatically.

A file system your agents can use

Your agents get the same simple tools to search, read, write and forget notes in your own memory: your profile, project notes, a scratch area. Claude Code, Codex, GitHub Copilot, Gemini CLI, Grok Build, Mistral Vibe and API agents alike.

Switch AI, keep your context

Your profile and the most relevant memories are handed to whichever agent you open, so switching from Claude to Gemini doesn’t mean explaining yourself again.

Private by construction

With a hub, your memory lives on your own computers, encrypted, and syncs only between your devices. No hub? You get a smaller version in your encrypted cloud space. One memory per person — never shared, never used to train AI.

Tidy, and it forgets when you ask

Delete a chat and what was learned from it is forgotten too. Memories that say the same thing are merged, and the memory is packed so it takes far less space.

You stay in control: it is on by default, and you can switch it off or clear it any time in Settings → AI agents → Agent memory. Read how Memory OS works →

For hub users

Video maker — ask your agent for a video, get a file back

Tell any of your agents “make a 60-second video about…”. It writes the script, and your own computer turns it into a finished video: your voice settings read it aloud, your own clips and pictures fill the screen, and the words appear as subtitles.

Your clips, or pictures that are free to use

No stock-footage subscription and no guessing about licences. Use your own clips and images, or ask for a picture by description: it is found on Wikimedia Commons and similar open collections, only with licences that allow reuse (public domain, CC0, CC BY), and the credits are written for you.

Your own keys, nothing extra to sign up for

The script comes from the agent you already use, the voice from your voice settings (Piper, a Microsoft voice or your ElevenLabs key). No separate subscription and no account with a video service.

No social-media posting

KumbayAI never logs in to TikTok, Instagram or YouTube for you. You get an .mp4 in your chat and publish it yourself, if and where you want.

Made on your computer

Phone-ready 9:16, 16:9 or square, with optional background music of your own. Rendering happens on your hub; nothing is uploaded. It needs a hub with voice set up, which today means a Linux hub; macOS and Windows are next.

Why KumbayAI

The app no AI company will build.

Claude’s app has only Claude. ChatGPT’s has only ChatGPT. Gemini’s has only Gemini. Each company wants you in its own box. KumbayAI sits above all of them, so you can use the best AI for each question and never be locked in.

Ask them all at once

One question goes to several AIs together. You get a combined answer written by the AI you choose, who said what, and every full reply one click away. Disagree? Continue with whichever answered best.

Switch AI without starting over

Stuck with one AI, or is it down? Continue the same conversation with another, or start a fresh chat with a summary. No copying and pasting between tabs.

Every conversation in one place

Search all your chats from every AI with Ctrl+K, open two chats side by side (one writing, one reviewing), and see what each AI costs you.

Get started

Start in a minute. Go further when you want.

Most people · no install

Just your browser or phone

Start chatting with free AI services without installing anything. Create an account, pick Google Gemini, Groq, OpenRouter, NVIDIA or Mistral, get a free API key from their site and paste it. That’s it. Any paid key or other OpenAI-compatible service works too.

  • Chat, Ask All, Split Chat, search, saved prompts and voice, in any browser or on your phone
  • Chats, agents and keys encrypted with your account key, on every browser you sign in to
  • Install a hub later and move everything onto your own computers with one click

Power users · optional

Add your computers

Install the hub on Windows, macOS, Linux or a Raspberry Pi. Your agents then run there: Claude Code, Codex, GitHub Copilot, Grok Build, Mistral Vibe, Antigravity, local models — with access to your files and commands, scheduled prompts, backups and more.

  • Sign in with the subscriptions you already have, or with a free account
  • Everything stays on your devices (plus an encrypted cloud backup of your chats); the hub updates itself
  • Let claude.ai use your computers with the KumbayAI connector

Privacy by design

Never read. Never shared. Never used to train AI.

Your chats, files, API keys, logins and settings live on your own computers. Everything that travels between your phone or browser and your agents is end-to-end encrypted with keys that exist only on your devices. Our relay just forwards sealed traffic it cannot read — like a postal service carrying locked boxes it has no key to.

  • No readable chat history, prompts, replies or files on our servers. With a hub, your conversations stay on your computers, plus an encrypted backup of your chats in our cloud (you can turn it off).
  • No hub? Your chats are encrypted with your account key. In no-hub mode your browser encrypts your chats, agents and API keys with your account key before they are saved. We store only the locked boxes.
  • Simple sign-in, or Maximum privacy — your choice. To make sign-in simple, KumbayAI keeps a protected copy of your encryption key (your account key) and releases it only to you, after two-factor sign-in; you get an email when a new browser receives it. That copy means KumbayAI could technically access the key. Turn on Maximum privacy in your account and the copy is deleted: the key then exists only on your own browsers and computers, and a new browser gets in with your passkey, your recovery key or an approval from one you already use.
  • Nothing shared with or sold to third parties — no ads, no tracking, no data brokers
  • Delete your account and it is gone — its computers, encrypted chats, cloud backup, keys and sign-in records are removed from our servers
  • Never used to train AI models — not by us, and we never pass your data to anyone who could
  • Backups are encrypted on your computer — your chats in our cloud, everything in a folder on each computer

Two honest exceptions. (1) The KumbayAI connector lets claude.ai use your computers. The commands, file contents and results of those tool calls pass through our relay in transit — encrypted on the wire but not end to end, because claude.ai can’t hold your keys. We never store them. (2) A few AI services (such as NVIDIA) don’t accept requests from a web page. For those, your browser’s requests go through our relay’s pass-through, which forwards them in transit and never stores them. The app tells you when an agent uses it.

The one place your prompts go is the AI service you pick for an agent. Your computer or browser sends them there, signed in with your own account, so that provider’s terms and privacy settings apply. To run your account we keep the basics: your email address, your list of paired devices and their names, the connectors you allowed, and your plan. Details in the privacy policy.

Phone or browser you, anywhere
KumbayAI relay forwards sealed traffic
can’t read it
Your computer hub, agents, chats,
files & keys

Pricing

Try everything free for 28 days. Then pick a plan.

Every new account starts with 14 days of everything, then 14 more days of no-hub mode — no card needed. Prices include VAT. Cancel any time.

No-hub

$3 / month

or $30 a year — 2 months free

  • Chat in your browser or on your phone with your own API keys — free AI services work too
  • Ask All, Split Chat, search, saved prompts and voice
  • Chats, agents and keys encrypted with your account key
  • The KumbayAI connector for claude.ai

Hub

$5 / month

or $50 a year — 2 months free

  • Everything in No-hub, plus:
  • The hub on up to 5 computers: Windows, macOS, Linux or Raspberry Pi
  • Claude Code, Codex, GitHub Copilot and more, signed in with the subscriptions you already have
  • Files and commands on your computers, scheduled prompts, backups, health alerts and Wake on LAN

Prices in US dollars and include VAT; your local currency may be shown at checkout. Our reseller Paddle handles payments, invoices and taxes as the merchant of record. The AI services you use (API keys or subscriptions) are billed by their own providers, never by KumbayAI. When a trial ends nothing on your devices is deleted — the service pauses until you subscribe. See the terms and the refund policy.

Save on AI costs

Use the subscriptions you already pay for — or free ones.

Sign in to Claude or ChatGPT with the account you already pay for, instead of using a pay-per-token API key. Your agents then run on your existing plan’s allowance — no surprise token bills. Several AI companies also offer agents you can sign in to with a free account.

Pay-per-token API key

Every prompt and every file an agent reads is metered. A heavy coding day with an agent can easily run into tens of dollars in API tokens — and long, busy sessions more than that.

Sign in with your subscription

The same work is covered by a plan you may already have, within that plan’s usage limits. No extra per-token cost. Claude Code needs a Claude Pro or Max plan; Codex CLI needs ChatGPT Plus or higher.

Free sign-ins and free keys

Start free in one click: sign in with OpenRouter and chat with its free models, no key to copy. Free means small: about 50 messages a day, and the free models’ providers may log your prompts and use them to train their models, so keep private information out of them.

Antigravity, GitHub Copilot Free, Grok Build and Mistral Vibe run on the hub with a free account. Google Gemini, Groq, NVIDIA and Mistral give free, rate-limited API keys that also work without a hub. Every free key you add, OpenRouter’s too, joins one “Free AI” agent that uses the one with the most room left.

Illustrative only — actual costs depend on the model, provider pricing and how much you use it. Free plans and free keys have limits set by each provider, and their terms apply. API keys are fully supported too, and KumbayAI’s usage tracking shows what each agent costs. KumbayAI never resells AI usage or adds a markup.

And there’s more

Also in KumbayAI

Stand-in agents — when an AI service is down, keep going

If a reply fails because the service itself is in trouble — down, overloaded, out of its usage for hours, or your sign-in there expired — KumbayAI offers your other agents that can answer right now. Pick one and it continues in a new chat with the conversation so far and the message that failed, which it answers at once. No need to repeat anything, and Agent memory works as in any chat.

KumbayAI watches the service for you for up to 24 hours — with a hub, even after you close the browser. When it’s back you get a note: “Claude is back online — do you want to switch back?” Switch back, and everything said in the meantime is copied into the first chat, so the first agent knows it too.

KumbayAI in its own window

On your Mac or Windows PC, “Open KumbayAI” in the menu bar or tray opens the app in a window of its own, with a Dock or taskbar icon. Close it and KumbayAI keeps running in the menu bar or tray. Signing in with Google happens in your browser: check that the 6-digit code matches, and the window signs in by itself.

Free AI — one agent for every free key

Free keys from Google Gemini, Groq, NVIDIA, Mistral and OpenRouter all join one “Free AI” agent. In the background it balances their rate limits, daily quotas and how long a chat each can take, and the line under each reply says which service wrote it.

Split Chats — up to 4 agents on one screen

Tick up to four chats — from the same agent or any mix of agents — or start new ones, and the screen splits into live panes. Each pane is a full chat with its own agent, device, attachments and voice. Claude Code refactoring in one pane, Codex reviewing in the next, Gemini researching in the third: watch them all at once.

Send to all splits: type once and every pane gets the message. Resize the panes by dragging, and save a layout as a workspace to reopen it in one click. Pop up: move the whole split into its own window outside the browser, for example onto a second monitor. The agent list shows an orange blinking light while an agent is working on your prompt and a green one when a new reply is waiting, so you never have to check.

Up to 4 side by side on a computer, 2 on a tablet, swipeable tabs on a phone; Pop up is for computers and larger tablets.

With the hub

Health monitoring that investigates for you

Got a home server, a NAS or a few computers? Health monitoring is on by default: KumbayAI watches CPU and drive temperatures, disk space, drive health (S.M.A.R.T.), Raspberry Pi power problems, and runs a nightly check of security and system events. When something goes wrong you get an alert in the app and a push notification, even when KumbayAI isn’t open. Then “Investigate with AI” reads the logs on that computer and tells you the real cause in plain words — and where there is a safe fix, “Fix this for me” shows you the exact commands and runs them only after you confirm. Both keep running on the computer if you leave the page or put your phone away, and you get a note when they’re done. You can switch monitoring off in Settings → Devices.

No-hub mode

Start chatting with free AI services without installing anything. Your chats are stored encrypted with your account key (with Maximum privacy, only your own browsers and phone can open them). Chat, Ask All, Split Chat, search and voice all work right in the browser, and a hub can take everything over later.

KumbayAI connector for claude.ai

Add KumbayAI as a custom connector in claude.ai — the Free plan too — and Claude can list folders, read and write files and run commands on your computers. You pick the devices, folders and what needs your OK, and you approve requests from the app, your phone or the Windows tray. Privacy: Claude’s tool calls pass through our relay in transit and are never stored.

Free sign-in agents

Antigravity, GitHub Copilot Free, Grok Build and Mistral Vibe with a free account, plus one-click set-up for free API keys from Google Gemini, Groq, NVIDIA and Mistral — all joined into one Free AI agent.

One sign-in for all your devices

One account key opens all your devices. Sign in on a new browser with your password and authenticator code and it’s ready. Prefer Maximum privacy? Then the key stays only on your own browsers and computers, and a new browser is unlocked with a passkey, an approval or your recovery key.

Ask All in one window

One message to several AIs, a combined answer written by the summarizer you choose (with who said what), and each agent’s full reply a click away. Continue with any of them.

Search all chats

Press Ctrl+K (⌘K on a Mac) to search all your chats by words or phrases, filtered by agent and date, and jump straight to the message.

Device groups & Wake on LAN

Group your computers, run scheduled prompts on a group, get told about missed runs, and wake a sleeping computer from the app. Wake on LAN is on by default.

Raspberry Pi hub

An always-on hub that draws a few watts: your agents, schedules and backups stay reachable while your laptop sleeps. Raspberry Pi 3, 4 and 5 (64-bit OS recommended).

Windows and macOS hubs

Windows 10 and 11 without administrator rights, running next to the clock, and a macOS installer for Apple chip and Intel Macs (beta). Both update themselves and open KumbayAI in its own window.

Multi-region relay

The relay now runs in Europe and the US. You connect to the nearest one, and if one region has trouble, traffic moves to the other by itself.

What you get

Everything you need to work with many agents

Multi-agent chat

Claude Code, Codex, Antigravity, GitHub Copilot, Grok Build, Mistral Vibe, OpenRouter, API-key services and local models in one app, with live streaming replies.

Ask All Agents

Send one question to several AIs at once. KumbayAI gives you one combined answer, with every full answer underneath to compare, and you carry on with the one you like best. In the app it sits right under Split Chats.

Split Chats

Run up to four chats side by side on one screen, from the same agent or any mix. Send one message to all of them, save a layout as a workspace, or pop the split out into its own window. The agent list blinks orange while an agent works and green when a reply is waiting.

Free AI

Add free keys from several services (Google Gemini, Groq, NVIDIA, Mistral, OpenRouter): they become one agent, not one each. For every message it picks the service with the most room left under its rate limits and daily quota, and one that takes a chat that long. If one fails, the next one answers, and the reply tells you which. You hit the free limits far less.

Saved & scheduled prompts

Keep your favourite prompts at hand, or have an agent run a task every morning and notify you when it’s done.

Context tracking

See how full each agent’s context is. When it gets long, one click summarizes the chat in the background and carries on in the same chat with a clean context, marked by a blue line. No duplicate chats, and nothing is forgotten.

Attachments

Send images, videos and other files up to 50 MB to your agents — with a hub, they go straight to your own computer. And every file an agent makes for you (a PDF, a script, a video, a picture) comes back in the chat, so you never have to go looking for it on a computer.

Voice in, voice out

Dictate prompts and hear answers read aloud with Piper, ElevenLabs or your browser’s voices — summarized, without reading out paths and links.

Health alerts you can act on

The Critical / Warning / Healthy counts are buttons: click one to see every alert of every device in one window. Acknowledge an alert and it clears everywhere — the top bar, the Health overview, the Devices tab and the bell — or snooze it. Installs, updates and a new device’s first setup don’t raise false “new account” alarms.

Clean up (hub, Linux today)

Find large files, exact duplicates and leftovers that are safe to clear, on any device with a hub. It scans first and removes nothing until you pick the items and confirm. Files go to the Trash so you can get them back, and every action is logged.

Video maker (hub, Linux today)

Ask an agent for a video: it writes the script, your computer records the voice, adds subtitles and builds it from your own clips, from pictures you describe, or from pictures your agent made. No stock footage and no social-media posting; you get an .mp4 in the chat.

Pictures by description, found or AI-drawn

“Show a yellow tram in Lisbon”: the video maker finds an openly licensed photo (Wikimedia Commons and similar collections; public domain, CC0 or CC BY only) and writes the credit lines for you to paste when you publish. Or ask for a picture to be drawn by AI, for a video scene or on its own, with a free picture model (FLUX) that needs no account or key; only your description is sent. If that service is busy, a second free one steps in.

Usage & cost tracking

Know what each agent and model costs you, per day and per device.

Notifications

Get told when a long task finishes or a scheduled prompt has run.

Encrypted backups

Encrypted backups: your chats in our cloud, chats and settings in a folder on each computer.

Import your history

Bring in chats from Claude Code, Codex CLI, Gemini CLI, Cursor, Copilot, LM Studio, Agent Hub, Hermes and OpenClaw, and exports from claude.ai, ChatGPT, Gemini and Grok. Imported chats are added to your agent memory, so your agents know what you told the other app.

One sign-in, all devices

Name (and rename) each computer and pick which agents run where. Pair a new computer by scanning a QR code or typing the short code it shows. Removing a device from the app also uninstalls the hub from it. One account, protected by two-factor authentication.

Automatic updates

The hub and the agent programs it runs keep themselves up to date.

How it works

  1. Create an account — protected by two-factor authentication.
  2. Start chatting right away with a free AI service in your browser, or install the hub on a computer and link it to your account:
    • Linux (Debian/Ubuntu, Fedora/RHEL, openSUSE; x86-64 and ARM64) and Raspberry Pi: 
      curl -fsSL https://kumbayai.com/install.sh | sudo sh
    • Windows 10 or 11: download the installer and open it — no administrator rights needed. It runs next to the clock.
    • macOS: download the installer for a Mac with Apple chip or Intel. During the beta it isn’t signed yet, so macOS asks you to allow it once; the app shows you how.
    • Docker: run the hub as a container — docker run -d --name kumbayai-hub --restart unless-stopped --init -v kumbayai-data:/data -v kumbayai-workspace:/workspace kumbayai/kumbayai:latest, then open the link in docker logs -f kumbayai-hub. The app has the full command and a compose file (Devices → Add a computer → Docker).
  3. Use your agents from anywhere — the web app talks to your hub over an end-to-end encrypted connection.

Coming soon

On the roadmap

Frequently asked questions

Can KumbayAI read my chats?

We don’t. With a hub, your chats, files, keys and settings are stored on your own computers, and traffic between your phone or browser and your hub is end-to-end encrypted; our relay forwards it without being able to read it. Without a hub, your browser encrypts your chats, agents and keys with your account key before they are saved on our servers. To make sign-in simple, KumbayAI keeps a protected copy of that account key and releases it only to you after two-factor sign-in — so, technically, KumbayAI could access it. If you want KumbayAI to be unable to access it at all, turn on Maximum privacy (Account page): the copy is deleted and the key exists only on your own browsers and computers. The other exceptions are the KumbayAI connector and the pass-through (see the next questions): there, data passes through our relay in readable form while in transit, and is never stored.

Do I need to install anything?

No. You can start chatting in your browser or on your phone with a free API key from Google Gemini, Groq, OpenRouter, NVIDIA or Mistral. Agents that are programs — Claude Code, Codex, GitHub Copilot, Grok Build, Mistral Vibe, Antigravity — and local models need the hub on a computer, and so do scheduled prompts, computer control, device groups and Wake on LAN. Once you install a hub, KumbayAI offers to move your chats and agents onto it.

What is the KumbayAI connector?

A way for claude.ai — including the Free plan, which allows one custom connector — to use your computers through your KumbayAI hub. In claude.ai, add https://relay.kumbayai.com/mcp as a custom connector and sign in to KumbayAI. You choose the devices and folders and whether reading, commands and writing files are allowed, need your OK each time, or are off; Account → Connectors → Disconnect removes it. Privacy: claude.ai (Anthropic) is the one asking, so the commands, file contents and results of its tool calls can be read by Anthropic and pass through our relay in transit (encrypted on the wire, not end to end). KumbayAI never stores them.

What is the pass-through?

Some AI services (such as NVIDIA) don’t accept requests sent from a web page. When a browser agent uses one of them, the request goes through our relay, which forwards it to that service and streams the answer back. It is readable to the relay while in transit, and never stored or logged. The agent’s settings show whether it runs “directly from this browser” or “through KumbayAI’s pass-through”. Agents on a hub never use it.

So what does KumbayAI store?

What’s needed to run your account: your email address and sign-in details, your two-factor registration, the list of devices you’ve paired (with the names you gave them and when they were last seen), the protected copy of your account key that makes sign-in simple (unless you turned on Maximum privacy) and a short list of the browsers it was released to, locked copies of your account key that only your passkey or recovery key can open, the connectors you allowed and their settings, your plan, and short-lived security logs. Without a hub, also your encrypted chats, agents and keys. See the privacy policy.

Should I sign in with my subscription or use an API key?

If you already pay for Claude or ChatGPT, signing in with that account usually costs you nothing extra — the agents use your plan’s allowance. Claude Code needs a Claude Pro or Max plan, and Codex CLI needs ChatGPT Plus or higher. Antigravity, GitHub Copilot Free, Grok Build and Mistral Vibe work with a free account, within their limits. API keys are billed per token, which suits light or occasional use, specific models, or going beyond your plan’s limits. KumbayAI supports all of them, and you can mix them per agent. Each provider’s own terms apply.

Which devices can I use?

The hub — the part that runs your agents — installs on Linux (Debian/Ubuntu, Fedora/RHEL and openSUSE, x86-64 and ARM64), on a Raspberry Pi 3, 4 or 5, on Windows 10 and 11 (64-bit), and on macOS (Apple chip and Intel; beta installer). You use KumbayAI from any modern browser on desktop or phone; a dedicated phone app is coming.

What happens if KumbayAI’s servers are down?

Your chats, history and agents stay safe on your own devices — nothing is lost. You won’t be able to reach your hubs remotely until the service is back, but the agents keep running and scheduled prompts keep working on your computer. Without a hub, your browser keeps a local copy of your chats and syncs it again when the service is back.

Do I need to open ports on my home network?

No. Your hub makes an outgoing connection to the relay, so there’s nothing to forward or expose.

How does the free trial work?

Every new account gets 14 days of everything, including hubs on your computers, then 14 more days of no-hub mode in the browser. No card is needed. After that the service pauses until you choose a plan — nothing on your devices is deleted, and you can subscribe any time to continue where you left off.

Can I cancel, and do I get a refund?

Yes, any time, from Account → Plan or from the link in your Paddle receipt. Your plan runs until the end of the period you paid for. In the first 14 days after your first payment you can ask for a full refund. Details in the refund policy.

How do I join the beta?

The beta is invitation-only for now. Write to privacy@kumbayai.com if you’d like an invite.

Every AI. One app.

Start chatting free