iPhone + Mac companion · Claude Code today
Vyber turns every permission prompt, question, and finished diff from your Claude Code sessions into a lock-screen tap. No SSH. One command to connect.
$ npm i -g vyber-companion && vyber connect
The whole loop: under 10 seconds. From your lock screen.
The problem
You've got worktrees. Subagents. A Max plan doing real work across three projects at once. Each session runs for twenty minutes, then blocks on one thing: a permission prompt, a question, a finished diff.
So you babysit terminals. Cmd-tab, scroll, squint, approve, repeat — or you walk away and come back to five agents that have been sitting idle since the moment you left.
The bottleneck isn't your agents. It's that their interrupts can't reach you. Vyber routes every one of them to the device that's already in your pocket.
The cockpit
One ranked queue for everything that needs a human: permission prompts, questions, finished diffs, stuck agents. Each item carries its context — the command, the agent's reason, the file list — so you can resolve it in one tap instead of one terminal hunt.
Every session across every paired Mac: project, branch, state, what the agent is doing right now, and how long it's been at it. Glance, confirm all green, put the phone back in your pocket.
When an agent finishes, the diff comes to you: syntax-highlighted, mobile-optimized, file by file. Approve it, or send it back with a steering prompt — "add a test for the nil case" — without opening your laptop.
Burn tracking
Vyber watches your sessions' token burn and projects it forward. Running three Opus sessions at 2pm? You'll know you're on pace to hit the wall at 3:40 — in time to move the grunt work to a cheaper model, not after the lockout.
Best-effort, from what's observable in your local transcripts. Estimates, not billing data.
Setup
01
$ npm i -g vyber-companion
$ vyber connectA lightweight daemon plus a Claude Code plugin. It hooks session events — it never touches your terminal workflow.
02
vyber connect prints a short-lived pairing code in your
terminal. Scan it with the iPhone app. Multiple Macs, one account.
03
Next time an agent needs you, your phone buzzes with the context and the buttons. That's the whole product.
Where Vyber fits
Anthropic's Remote Control and the open-source mobile clients are good at what they do: a chat with a session, from anywhere. Vyber works alongside everything Anthropic ships — it's for the moment you stop having "a session" and start having a fleet.
| Vyber | Claude app Remote Control | Open-source clients | |
|---|---|---|---|
| Interaction model | Ranked interrupt queue across all sessions | Chat transcript per session | Chat transcript per session |
| Built for | Triaging 2–10 parallel agents in seconds | Working a session from your phone — free and official | Self-hosting, E2E encryption, tinkering |
| Lock-screen actions | Approve / Deny / Reply from the notification | Push notifications; act in the app | Varies by project |
| Fleet overview + Live Activity | Fleet board, Dynamic Island agent status | Session list | Varies by project |
| Mobile diff review | Per-session, approve or request changes | In the conversation | Varies by project |
| Usage burn tracking | Projected limit alerts before you hit the wall | — | — |
Honest note: run one agent at a time? Remote Control is excellent and costs nothing extra. Vyber earns its keep at two-plus. Full comparison in the blog post.
Pricing
$0
The single-session experience, forever.
For fleet runners
$9.99/mo
or $79.99/yr — two months free.
FAQ
By default: session state (working / needs you / done), short activity summaries, permission requests, and your decisions. Your source code stays on your Mac. If you enable diff review, the diffs you choose to review sync over TLS so you can read them on your phone — that's the only time file contents leave your machine, and you can turn it off. The companion makes outbound connections only: no SSH, no port forwarding, no inbound ports, nothing listening.
Yes. Vyber sits on top of your existing Claude Code setup — it doesn't proxy your Anthropic account or consume your usage. Whatever plan runs your agents, Vyber relays their interrupts. It's built for exactly the people running Max plans with multiple worktrees in parallel.
Very little, by design. It's a small Node daemon that reacts to Claude Code hook events and tails transcript files — it doesn't poll your CPU, screen-record, or proxy your traffic. Sync traffic is state changes, not token streams: a few small writes per minute per active session. If your Mac sleeps, sessions show as offline until it wakes.
Nothing breaks. Interrupts queue and your terminal keeps working exactly as if Vyber weren't installed — a permission prompt you don't answer from your phone is still sitting right there in the terminal. When the connection returns, state syncs and pending interrupts reappear. The companion never blocks or times out your local session on Vyber's account.
On the roadmap. Vyber's companion is built on an adapter interface — Claude Code is the first adapter, and it's the deepest because Claude Code's hooks make clean interrupt detection possible. Codex and Gemini CLI adapters are planned; join the waitlist and tell us which one you need first.
You can — plenty of us did, for a while. But SSH from a phone gets you a tiny terminal you have to poll by hand: open the app, reconnect, resize, scroll five sessions to find the one that's blocked. It also means exposing an inbound port or running a tunnel. Vyber inverts it: the interrupt finds you, with context and buttons, on the lock screen. Ten seconds instead of two minutes, and nothing on your Mac is listening.
Waitlist
iPhone-first, invites in waves. Tell us how many agents you run — heavy parallel users get bumped up the list.