שָׁלִיחַ · shaliach · n. one you send to act in your name
ShaliachAI agents that run your company
Send one. A whole team goes.
Your own team of Claude Code agents, on your Mac or server. You text one, it hires the rest, and nothing goes out without your tap.
Runs 4 real businesses today: see the receipts.



Screens shown with demo data. See the receipts.
Shaliach for Business
A paid product, sold to businesses only. We install it on your Mac or server and do the first setup with you. Setup from $5,000, installed for you; we quote your exact scope on the call.
Windows too: Shaliach for Windows · Videos
The job description is two thousand years old
"A person's emissary is as the person himself."Mishnah, Berakhot 5:5
That is a lot of trust to hand to software. So ours carry your name, do the legwork, and still come back to ask before they sign anything.
How it works
- 1. You text Atlas.From the iPhone app, the Mac app or the web board. One line is enough: "Get me more customers this week."
- 2. Atlas hires the team.It picks the agents the job needs, briefs each one, and folds their chatter into one thread you can read.
- 3. Each one gets its own computer.A browser and a terminal per agent. Watch any screen live and take the wheel when you like.
- 4. Anything risky waits for your tap.Publishing, paying, logging in: Allow once, Always, or Not today, from your phone.
Want a team like this for your business?
We install it for you and do the first setup with you.
Meet the team
Atlas
Hires, briefs and turns forty messages into one you can read. Will still ask before anything leaves the building. Yes, even that.
Web
Writes the post in its own browser, shows you the preview, then waits for your tap like it's holding a door.
Social
You said "never post before 7am" once. Social wrote it down. So did everyone else.
Growth
Found 20 offices near the café that order lunch. Drafted 20 emails. Sent none until you said so.
Chaser
Polite on the first reminder. Still polite on the fifth. Frankly, more patient than you.
These are the agents from the app's built-in demo, a made-up café. Yours are whoever Atlas hires, and each one can pick its own shape and colour.
What your emissaries actually do
You text one. It hires the rest.
Atlas, your chief of staff, hires the agents the job needs, briefs each one and folds their back-and-forth into one thread you can actually read.
Their own computers, not yours.
Every agent gets its own browser and terminal. Watch any screen live, and take the wheel whenever you like.
Nothing goes out without your nod.
Publishing, paying, logging in: it stops and asks on your phone. Allow once, always, or not today.
Tell them once.
Correct one agent and the whole team writes it down. "Never post before 7am" stays never.
They come back with the goods.
Videos, PDFs and screenshots land in the chat, not in a folder you will never open again.
On your turf.
Your Mac or your own server, your Claude plan. Shaliach never stores your Claude token.
What a shaliach is not
- Not a chatbot with opinions. It doesn't tell you what you could do. It opens a browser and does it.
- Not someone else's cloud. It runs on a machine you own, under your own Claude account.
- Not unsupervised. Anything risky waits for your tap, from the Mac app or the iPhone app.
- Not a black box. Watch any agent's screen live, and read what the whole team did in one thread.
Questions people ask
What is Shaliach?
Shaliach is a team of always-on Claude Code agents that runs on your own Mac or server. You text one agent, Atlas, the chief of staff; Atlas hires the agents the job needs, gives each one its own browser and terminal, and asks you on your phone before anything risky happens. There is a Mac app, an iPhone app and a web board.
Can I run Claude Code agents from my phone?
Yes. You chat with Atlas and the team from the iPhone app (or the web board in any browser), approve risky steps with one tap, and watch any agent's screen live. There is no Android app yet.
Where do my messages, code and files go?
To your machine and nowhere else we run. Shaliach has no cloud of its own. Threads, files and agent workspaces live on the server or Mac we install it on. The agents talk to Anthropic the same way Claude Code always does, under your account. Voice calls are optional and go to OpenAI only if you add your own API key.
Does it touch my Claude login? Is this within Anthropic's terms?
For always-on or business use, use your own Anthropic API key, under Anthropic's commercial terms. Every agent is the unmodified claude CLI. You can also sign in with Claude's own claude auth login; the CLI keeps and refreshes its login, and Shaliach never stores your Claude token. Nothing is resold or pooled.
What will it cost me?
Shaliach is a paid product, for businesses only. Book a demo for a quote. On top of that, the agents use your own Claude plan, exactly like Claude Code sessions you start yourself. Five agents working at once use it about five times as fast as one. Start with one or two agents and watch your usage on the first day. We have not published a "typical cost" because we have not measured one we trust.
Which platforms work?
We install it for you on one of these. Server: Ubuntu 24.04 or Debian 12 (x86_64 or arm64), or macOS. Mac app: macOS 14+, Apple silicon or Intel. iPhone app: yes. Any browser: the web board. Windows 10 or 11: yes. There is no Android app yet.
What can the agents do on my machine?
Anything their Unix user can do. They run Claude Code in bypassPermissions mode so they do not stall on prompts. The approval cards (Allow once / Always / Deny) stop risky steps, but they are a convenience, not a sandbox. On a server with agent desktops on, the service user is in the docker group, which is root-equivalent. So use a VPS that does nothing else. The Mac bridge, which lets agents run jobs on your Mac, is off until you turn it on, logs every job, and can be stopped from the menu bar.
What is exposed to the internet, and how is it authenticated?
The deck binds 127.0.0.1 only. On a public server, Caddy passes only /v1 and /healthz over HTTPS on 443, and ufw allows only SSH and 443. Every /v1 call needs a bearer token, compared in constant time; with no token set, the API refuses everything. Each paired phone or Mac gets its own token, stored only as a hash and revocable. Wrong guesses are rate limited. Want nothing public at all? We can set it up behind an SSH tunnel or Tailscale. Secrets are plain files with 0600 permissions, not encrypted at rest.
Is it open source?
No. Shaliach is a closed, paid product for businesses. We install it for you; you get the agents, the apps and support.
Did an AI write this?
A lot of the code, yes: most of it was written by Claude Code agents, many of them running inside Shaliach, with a human designing, reviewing and testing every change. If something here reads wrong, write to support@shaliach.me and it gets fixed.
How is this different from OpenAI dots, Meta Muse or Grok Bot?
Those rent you an agent in their cloud, with no setup. Shaliach is a team of agents on hardware you own, on the Claude plan you already pay for. We install it and do the first setup with you. The trade-off is real: it runs on a machine you own and look after.
Is this affiliated with Anthropic?
No. It is an independent project that runs Claude Code. "Claude" and "Claude Code" are Anthropic's trademarks, used here only to describe compatibility.