Everything abs does, in one place

From the first ping to the way it goes quiet when you sit down — here's the full picture, and how the loop actually works.

Everything you need from your phone

New to this? Start with how the Telegram integration works — the flow, the setup, and what is safe about it.

Notifications

A written summary the moment a task finishes — not just a ping.

Remote replies

Your answer goes into the session that is already running.

Screenshots

Attach an image and Claude reads it directly.

Voice, both ways

Speak and be spoken back to — transcribed on your own machine.

Usage

Your Claude limits from the chat, and live in the status bar.

Multiple projects

One bot each, so their conversations never mix.

Personas

Pick who you are working with at launch. Four ship; write your own.

The prompt in your hands

Read and edit every layer abs puts in front of Claude.

Commits stay yours

Nothing is committed or pushed without you saying so.

Stay in control

Five phrases, sent from Telegram as a whole message. A session hook acts on them before the model does, so they work even if a message tries to talk it out of them.

Auto-silent

Reports mute themselves while you are at the terminal, and resume when you step away.

Mute

One message pauses the proactive reports. Inbound keeps working.

Stop

Halts the plan at the next step and waits for you.

Disconnect

Cuts Telegram in and out entirely; the session keeps working locally.

Exit

Closes the session, confirming first if it is mid-task.

Block

Locks the bot out until you re-pair at the terminal.

See the full command reference →

What it tells Claude, in your hands

Anything bolted onto a Claude Code session shapes how the model behaves, so none of it is hidden. abs prompt opens a tabbed page over every layer — the persona, the wording of each remote command, your CLAUDE.md, and what Claude remembers. Edit what is yours; reset any of it with one key.

Four personas ship

abs, ceo, cto and friend — or write your own.

Seven tabs

Persona, hooks, project and global CLAUDE.md, memory, and the locked mechanics.

Safety stays put

A persona changes the voice, never the rails — those are appended after it.

How the prompt page works →

It knows when you're at your desk

No pings while you're working — reports come back the moment you step away. You never flip a switch.

  • Silent while you work. A few prompts at the terminal and it stops pinging you.
  • Back on when you leave. Message it from Telegram, or just walk away — reports resume on their own.
  • The bar says who you are talking to. The bot, the persona you picked, and your limits — right in Claude's own status bar.
  • Usage at a glance. Your 5-hour and weekly limits, in the bar and on every report.
Claude Code
> refactor the auth module — running…
esc to interrupt · ⇧⇥ auto-accept edits
abs@pranBot · 🎭 cto · 5h 1% · week 8% · v3.7.4
Who it talks to, which persona is answering, and what is left of your limits. A state marker appears only when something is off.

How it works

1

You run abs instead of claude.

Agent Babysitter starts Claude Code with Telegram connected.

2

Claude works as usual.

It runs your tasks in the same live session.

3

You get notified.

When a task finishes, you get a summary on Telegram.

4

You reply, send images, or ask.

Your message is delivered back to the same session.

See exactly what abs adds, and where it lives in the code
Replies, images, voice, commands →
You
Telegram
Agent
Babysitter
Claude
Code
← Task summaries, notifications, usage

Open source. Built for the community.

Agent Babysitter is open source under the MIT License. Use it, improve it, share it.

Found a bug or have an idea?
Open an issue or submit a pull request.
v3.7.4  ·  MIT License  ·  About & contact  ·  Made with ♥ for Claude Code users