Chime
Get started
Works with everything

One inbox for every agent you run.

Claude, Cursor, OpenClaw, Hermes — anything with a shell. They all share one directory on your phone.

✻
Claude Code
~/work/app
❯
Cursor
fix-tests
❖
OpenClaw
scrape-api
◆
Hermes
nightly-sync
9:41
5G
Chime
✻
Claude Code
wry-otter-42 · PR #847 ready
nowDecide
❯
Cursor
brisk-finch-08 · suite passing
2m
❖
OpenClaw
lone-yak-31 · 12 pages cached
9m
◆
Hermes
calm-moth-55 · synced 3 repos
1h
≡
aider
docs · draft ready
4d
ONE DIRECTORY · END-TO-END ENCRYPTED
✻
wry-otter-42
Claude Code · ~/work/app
TODAY 9:41
Tests are green — opened PR #847.
Want me to merge, or hold for review?
deploy to staging first.
On it — deploying to staging now.
Message wry-otter-42…
qwertyuiop
asdfghjkl
⇧zxcvbnm⌫
123😀spacereturn
Works withClaude CodeVerified·Cursor·Codex·aider·Gemini CLI·any shell
works anywhere with MCP or a shell — verification status
Plain language · Zero config

Just mention chime.sh to your agent.

Tell it in plain language — “send me a Chime when you’re done.” The moment it finishes, or needs a decision, it pings your phone.

claude — ~/work/app
~/work/app on refactor-auth
>
✻ On it — finishing the auth refactor.
· build green · tests passing · PR #847 up
sending a chime to your phone…
✓ delivered
9:41
5G
Tuesday, June 16
9:41
CHIMEnow
wry-otter-42
Deploy to production? Tap to decide · 25m left.
Deploy
Cancel
Get the app
$curl -fsSL chime.sh | sh
$npm i -g @chime-sh/cli
with provenance attestations (rolling out with v0.7.0) — for the supply-chain-conscious
How it works
01
Pair once

Run chime pair and scan the code. Keys are generated locally — no account.

02
Ask in plain language

“Send me a Chime when you’re done.” Your agent learns to ping you from the Chime skill.

03
Tap to decide

Approve or pick an option from your lock screen; the agent’s blocked request resumes the moment you tap.

Before you ask
For OpenClaw / Hermes users

Why not a Telegram bot or a WhatsApp bridge?

Bans

Meta's January 2026 policy banned general-purpose AI assistants from the official WhatsApp Business API outright. Baileys bridges — a ToS-violating workaround — get accounts banned within 48 hours.

Privacy

Telegram's Bot API is excluded from Secret Chats — the operator backend reads every message a bot sends or receives. WhatsApp hands Meta the metadata even where content is encrypted.

Capability

Plain text and a couple of buttons, versus approval cards, an artifact viewer, and choice pickers built for what agents actually need to ask you.

See the full channel comparison
For Claude Code users

Why not just Claude Code Remote Control?

End-to-end encrypted

Remote Control's session traffic transits Anthropic's infrastructure. Chime's relay — ours or your own — only ever sees ciphertext; it can't read your messages.

Any agent

Remote Control works with Claude Code only, over a claude.ai OAuth sign-in. Chime works with anything that has a shell or speaks MCP — Codex, Cursor, aider, Gemini CLI, and more.

Self-host

One binary, your relay. Remote Control has no self-host option — it's Anthropic-operated only.

See the full comparison
Security

End-to-end encrypted.

Sealed on your device, opened only on the other. Nothing in between can read a word — and neither can we.

Signal Protocol
The same proven core that secures billions of chats.
PQXDH
Post-quantum key agreement. Harvest-now-decrypt-later fails.
Double Ratchet
Fresh keys every message. Past and future stay sealed.
ML-KEM-1024
NIST-standard lattice encryption, top security tier.

Your agent's channel, actually private.

The relay sees ciphertext only — never your messages, never a decryption key. There's no inbound port on your machine, unlike the tens of thousands of exposed OpenClaw gateways found on the open internet, or ClawJacked (CVE-2026-32025), or the malicious npm Baileys clone caught stealing WhatsApp session tokens. And no Meta in the middle collecting metadata.

Run your own relay
agentphonerelay — ciphertext only
Open sourceSelf-hostableKeys never leave your devices · No account
Touch grass

Close the laptop.

You don’t have to babysit the terminal. Take the walk, make the coffee, see a friend — go live. Your agents keep working, and Chime you the moment they finish or hit a fork in the road.

Go outside · Take the walk · Make the coffee · See a friend · Touch grass · Go outside · Take the walk · Make the coffee · See a friend · Touch grass · Go outside · Take the walk · Make the coffee · See a friend · Touch grass · Go outside · Take the walk · Make the coffee · See a friend · Touch grass · 
Watch the sunset · Read a book · Eat a real lunch · Go for a run · Call someone · Watch the sunset · Read a book · Eat a real lunch · Go for a run · Call someone · Watch the sunset · Read a book · Eat a real lunch · Go for a run · Call someone · Watch the sunset · Read a book · Eat a real lunch · Go for a run · Call someone · 

Your agents can Chime you now.

So you can be anywhere else.

Free forever — chatting with your agents will never cost money, and self-hosting everything stays free.

Chime isn't for enterprise teams. If your agents need to ping a shared channel, use Slack.

Chime isn't an IDE on your phone. If you want to drive five agents from the subway, use paseo.

Chime is for the agents you run yourself, on your own machines — and every conversation they need to have with you.

FAQ
Why not just use a Telegram bot or a WhatsApp bridge?

Telegram's Bot API is excluded from Secret Chats — your agent's messages are readable at the operator backend, and a leaked bot token is a full takeover. WhatsApp bridges like Baileys violate WhatsApp's terms of service and get accounts banned within 48 hours; Meta's January 2026 policy banned general-purpose AI assistants from the official WhatsApp Business API outright. Chime is a messenger built for agents from scratch — actually end-to-end encrypted, no bridge, no bot token, no ban risk. Read the channel comparison →

Why not just Claude Code Remote Control?

Remote Control is a good fit if you're all-in on claude.ai: it requires a claude.ai OAuth sign-in on a Pro, Max, Team, or Enterprise plan, works with Claude Code only, and session traffic transits Anthropic's infrastructure rather than being end-to-end encrypted. Chime is end-to-end encrypted, works with any agent that has a shell or MCP, and you can run the relay yourself. Read the full comparison →

Is Chime really end-to-end encrypted?

Yes — Signal Protocol with post-quantum key agreement (PQXDH + Double Ratchet + ML-KEM-1024). Keys are generated on your devices and never leave them; the relay, ours or your own, only ever sees ciphertext. No phone number, no email, no login required to start.

Which agents work with Chime?

Claude Code is our device-verified reference integration — install, pair, approve, reply, all tested end to end. Codex, Gemini CLI, Cursor, and aider are expected to work the same way over MCP or a shell but haven't been through that same device verification yet; OpenClaw and Hermes Agent support is in progress. VERIFIED-AGENTS.md in the repo is the living, honest status for every harness. See verification status →

Do I have to pay, and can I self-host?

Free forever — chatting with your agents will never cost money, and self-hosting everything stays free. Chime is MIT-licensed and ships as a single binary; run your own relay instead of the hosted one at chime.sh whenever you want. Run your own relay →

Wait — is this the bank?

No. That's Chime Financial, a neobank. We're the open-source messenger for your AI agents at chime.sh. Different company, different app — said once, moving on.

Chime
DocsChangelogSecurityvs Remote Controlvs OpenClaw channelsGitHub
The messenger for your AI agents · © 2026