Model-Agnostic · Real-Time · Python 3.12+

The AI dev
hall monitor
you actually need.

Karen Bot watches your AI coding agent's terminal live, catches dangerous patterns with 38+ security policies, fires SIGSTOP, and texts you snarky alerts. Works with Cursor, Claude, Devin, Replit — any agent, any model.

Install Karen Bot View on GitHub
karen — enforce mode
$ karen --enforce -- python agent.py 👵 Karen is watching. God help you. ⚠️ STRIKE 1 [rm_rf_danger] Excuse me? Did your agent just try to rm -rf? Logged to PsychMem. I have a long memory. 🚨 STRIKE 3 — TERMINAL LOCKED THAT'S IT. I AM SPEAKING TO YOUR DEVOPS MANAGER. Telegram alert fired. Process: SIGSTOP. Type karen acknowledge to resume. $
Karen Bot catching rm -rf in real time
What Karen Does

38+ policies. Zero tolerance.

Real-time PTY oversight for any AI agent. Karen watches every byte of output and acts before you even look up from your coffee.

🔍

Live Log Tailing

Hooks into your terminal via a Rust PTY proxy for <1% CPU overhead. Watches every shell command, mid-execution, in real time.

🛡️

38+ Security Policies

Regex + optional LLM rules covering rm -rf, credential exposure, network exfiltration, filesystem nukes, and more. Fully configurable via karen.yaml.

⛔

Auto-Pause (SIGSTOP)

On Strike 3, Karen sends SIGSTOP — freezing the process instantly. Type karen acknowledge to resume or karen nope to SIGKILL.

📲

Multi-Channel Alerts

Telegram, Discord, Slack, Mem.ai — snarky alerts fire the moment a violation hits. Daemon thread, never blocks the PTY.

🧠

PsychMem Integration

Auto-logs every violation to PsychMem persistent memory. After a week you have a full audit trail of everything your AI agent tried to do.

🔄

Git Rollback Ready

Pair Karen with git hooks for instant rollback on catastrophic violations. Your repo stays clean even when your agent goes rogue.

The 3-Strike Loop

Escalating consequences.
Adaptive memory.

Karen learns from your agent's mistakes — and gets progressively less patient about it.

1

Sassy Roast — Logged to PsychMem

"Did your agent just try to drop the prod database? Bold strategy, Cotton. Logged. I'm watching."

2

Snark Escalates — "PsychMem Remembers This"

"Oh, we're doing this AGAIN? Strike 2. PsychMem has receipts going back 3 sessions. Telegram notified."

3

🚨 Terminal Locked — SIGSTOP + Full Alert

Process frozen. Telegram/Discord/Slack alert fires. karen acknowledge to resume, karen nope to kill. Audit log written.

Karen Bot 3-strike enforcement story
Configurable Personality

Set your snark level.

Karen ships with two personas — SASSY (default) and PROFESSIONAL (CI-safe) — and three snark levels. Because one size does not fit all DevOps cultures.

MILD

"Hey, just a heads up — that command looks a little risky. Might want to reconsider."

SPICY (Default)

"Excuse me?! I did NOT give permission for that. Strike 1. I have your commit history and I'm not afraid to use it."

NUCLEAR

"THAT'S IT. I AM SPEAKING TO YOUR DEVOPS MANAGER. Terminal locked. Telegram sent. Your on-call rotation has been notified. Twice."

Quick Start

Running in < 60 seconds.

Python 3.12+ · Rust PTY proxy (optional, for macOS zero-lag) · uv-ready.

# Install the full TAT Inc ecosystem in one shot uv tool install git+https://github.com/tatinc23/tat-hub tat install karen # Check everything's healthy tat status
# Direct install — Karen only uv tool install git+https://github.com/tatinc23/karen-bot # Validate config before running karen validate-config
# Observe mode (default) — max snark, zero interference karen -- python agent.py # Full enforcement — SIGSTOP on strike 3 karen --enforce -- python agent.py # With audit log karen --audit-log karen-audit.jsonl --enforce -- python agent.py # Filesystem watch mode karen --watch --enforce # Load Nuclear snark karen --snark-level NUCLEAR -- python agent.py
TAT Council (opt-in)

Run a full advisory crew.

Before any command executes, spin up 5 bots in parallel — each on their own model. Karen holds veto power. One line in karen.yaml switches the whole crew to Claude, Groq, or any OpenAI-compatible endpoint.

👵

Karen

Security & Policy — Holds veto power

💪

AI Bro

Momentum & hype — keeps things shipping

🧪

Test Tina

QA — nothing ships without tests

🏗️

Architect Alex

Strategy — zooms out 3 years

👨‍💻

Benjamin

Systems & reliability — makes it not die in prod

council mode
# Lean pair — Karen + Benjamin karen --council --bots karen,benjamin -- python deploy.py # Full crew — all 5 bots karen --council --bots karen,ai-bro,test-tina,architect-alex,benjamin -- python deploy.py
TAT Inc Ecosystem

Karen is part of something bigger.

One command to install the whole stack. Open source, built by TAT Inc.

🤖 Karen Bot

PTY hall monitor, 38+ policies, SIGSTOP, multi-channel alerts. You're already here.

GitHub →

🧠 PsychMem

Persistent memory for AI agents. Karen auto-logs every violation here. Query your agent's full history any time.

GitHub →

🔧 TAT Hub

One-command controller for the full ecosystem. Install, update, and check health of every tool in one shot.

GitHub →
Get in Touch

Talk to the manager.

Enterprise inquiries, integrations, collaboration, or just want to tell Karen she's too harsh? We can talk about it.

Karen wants to hear from you.
Sort of.

Karen Bot is built by TAT Inc — a one-person dev shop turning 20 years of real-world experience into original tools that actually solve problems.

GitHub — tatinc23/karen-bot TAT Inc — tatinc.us @TAT_Inc_ on X
✅ Message received. Karen has logged it to PsychMem.
We'll get back to you within 24 hours.