diff --git a/README.md b/README.md
index 200b080..5da4129 100644
--- a/README.md
+++ b/README.md
@@ -26,7 +26,6 @@ graph TB
 
 
    subgraph bot[status-im/status-python-sdk]
-        REQUIREMENTS[requirements.txt]
         GROUP_CHAT[class GroupChat]
         SDK[class Account]
         SIGNAL[class Signal]
diff --git a/examples/group-chat-moderator/README.md b/examples/group-chat-moderator/README.md
new file mode 100644
index 0000000..0c0def7
--- /dev/null
+++ b/examples/group-chat-moderator/README.md
@@ -0,0 +1,104 @@
+# Group Chat Moderator
+
+An **automatic moderator** for a [Group Chat](https://status.app/help/messaging/create-a-group-chat). The script logs into a Status account, listens for new messages **in real time**, and scores each one for toxicity with [Detoxify](https://github.com/unitaryai/detoxify) (local model). Authors of toxic messages are **warned**, and after a specified number of warnings they are **removed** from the chat.
+
+## How it works
+
+Every message that lands in the chat is handled in its **own thread** by [`check_message`](./main.py). The thread:
+
+1. Skips messages sent by the bot itself.
+2. Scores the message text with [Detoxify](https://github.com/unitaryai/detoxify) and takes the highest label (`toxicity`, `insult`, `threat`, ...).
+3. Ignores anything below the `threshold` (default `0.6`).
+4. Otherwise increments the author's warning count under a lock - the `warnings` dict is shared across threads, so the read-modify-write must be atomic.
+5. Sends a warning reply, or [`remove`](../../docs/group-chat.md#removepublic_keys)s the author once they hit the `warning_limit` (default `3`).
+
+```mermaid
+sequenceDiagram
+    actor Member as Group Chat Member
+    participant Listen as listen_messages
+    participant Check as check_message thread
+    participant Model as Detoxify model
+    participant Warnings as warning counts
+
+    Member->>Listen: sends message
+    Listen->>Check: spawns per-message thread
+    Check->>Model: score message text
+    Model-->>Check: highest label + score
+
+    alt Below Threshold
+        Check-->>Check: ignore
+    else At / Above Threshold
+        Check->>Warnings: increment author's count
+        Warnings-->>Check: current count
+        alt count < warning_limit
+            Check->>Member: send warning reply
+        else count >= warning_limit
+            Check->>Member: remove from chat
+        end
+    end
+```
+
+
+This is just one moderation policy. [`check_message`](./main.py) is self-contained, so you can rewrite it for your own use case - swap in a different model or keyword filter, adjust `threshold` and `warning_limit`  or escalate through different labels. The listener loop stays the same and only the per-message logic changes.
+
+**Note**: Removing members requires the account to be the [administrator](../../docs/group-chat.md#administrator) of the chat. See [Moderation power](./README.md#moderation-power).
+
+## Setup
+
+### 1. Install
+
+From the **repository root**, install the SDK with the `group-chat-moderator` dependencies:
+
+```
+pip install ".[group-chat-moderator]"
+```
+
+This pulls in [Detoxify](https://github.com/unitaryai/detoxify) and its [PyTorch](https://pytorch.org/) backend. The first run downloads the model weights.
+
+
+**Note**: `detoxify` installs the CPU build of PyTorch by default. For faster inference on a CUDA GPU, uninstall `torch` and `torchvision`, then reinstall the GPU builds by following the instructions on [PyTorch's website](https://pytorch.org/).
+
+### 2. Configure
+
+Copy [`env.example`](./env.example) to `.env` in this folder and fill it in:
+
+```
+cp env.example .env
+```
+
+| Variable | What it is |
+|-----|-------------|
+| `PASSWORD` | The password of the moderating Status account. |
+| `NAME` | The [display name](../../docs/account.md#display-name) or ENS name of the account. If you have previously logged in with the SDK you can provide an ENS. For first time log ins, it is best to provide a [display name](../../docs/account.md#display-name). |
+| `MNEMONIC` | The [recovery phrase](https://status.app/help/profile/understand-your-status-keys-and-recovery-phrase) of the account. Used to recover it into the container. |
+| `GROUP_CHAT_ID` | The `id` of the group chat to moderate. Group chat IDs come from the [`chats`](../../docs/account.md#chats) property, where `type` is `group_chat`. |
+
+### 3. Run
+
+The script loads its `.env` from the current directory, so run it from inside this folder:
+
+```
+cd examples/group-chat-moderator
+python main.py
+```
+
+On the first run, [`launch_docker_container`](../../docs/utils.md#launch_docker_container) builds the Status Backend image, which takes a few minutes. Tthe bot starts listening:
+
+```
+[INFO]  Successfully logged in!
+[INFO]  Loading Detoxify [cpu]
+[INFO]  Listening Group Chat Status Bots
+```
+
+Detoxify runs on the **GPU** automatically when CUDA is available (`[cuda]` above), and falls back to the CPU otherwise. The bot runs until you stop it with `Ctrl+C`.
+
+## Moderation power
+
+**This account acts as the moderator of the group chat.** To warn members it only needs to be in the chat, but to **remove** them it must be the [administrator](../../docs/group-chat.md#administrator) - only the admin can remove members. Point the moderator at a chat it created (or was made admin of), otherwise removals are rejected and members can only be warned.
+
+The moderation logic in this example is deliberately simple:
+
+- **One model, one threshold.** Every message is scored by [Detoxify](https://github.com/unitaryai/detoxify); anything scoring `0.6` or higher on any label counts as toxic. Tune `threshold` and `warning_limit` in [`check_message`](./main.py) to make moderation stricter or more lenient.
+- **Warnings are per public key.** The count lives only in memory, so restarting the bot resets everyone's warnings to zero.
+
+Detoxify is a machine-learning model and will make mistakes - both false positives and false negatives. Treat it as a first line of moderation, not a final judge.
diff --git a/examples/group-chat-moderator/env.example b/examples/group-chat-moderator/env.example
new file mode 100644
index 0000000..32ac755
--- /dev/null
+++ b/examples/group-chat-moderator/env.example
@@ -0,0 +1,7 @@
+# Status Account setup
+PASSWORD = "your-password-here"
+NAME = "status-display-name"
+MNEMONIC = "phrase_1 phrase_2 phrase_3 phrase_4 phrase_5 phrase_6 phrase_7 phrase_8 phrase_9 phrase_10 phrase_11 phrase_12"
+
+# Monitor messages from public chat
+GROUP_CHAT_ID = "group-chat-id"
diff --git a/examples/group-chat-moderator/main.py b/examples/group-chat-moderator/main.py
new file mode 100644
index 0000000..9875d0a
--- /dev/null
+++ b/examples/group-chat-moderator/main.py
@@ -0,0 +1,64 @@
+from dotenv import load_dotenv
+from status_sdk import Account, GroupChat, launch_docker_container
+from detoxify import Detoxify
+import os, threading, torch
+
+# `warnings` is shared by every check_message thread, so the
+# read-modify-write of a member's warning count must be atomic
+warnings_lock = threading.Lock()
+
+def check_message(account: Account, message: dict, warnings: dict, group_chat: GroupChat, model: Detoxify, threshold: float = 0.6, warning_limit: int = 3):
+    """
+    Score a single message and warn (or remove) its author.
+    """
+    public_key = message["from"]
+    if public_key == account.info["public_key"]:
+        return
+
+    label, score = max(model.predict(message["text"]).items(), key=lambda item: item[1])
+    account.logger.info(f"Message: '{message['text']}'\t\t{label} - {(score * 100):.2f}%")
+
+    if score < threshold:
+        return
+
+    with warnings_lock:
+        warnings[public_key] = warnings.get(public_key, 0) + 1
+        count = warnings[public_key]
+
+    if count < warning_limit:
+        group_chat.send_message(f"Warning {count} /{warning_limit} - @{public_key} please keep it civil.", message["id"])
+        account.logger.info(f"Sent warning to {public_key}")
+    elif count >= warning_limit:
+        group_chat.send_message(f"Removing @{public_key} member after {warning_limit} warnings.")
+        account.logger.info(f"Removed {public_key} from {group_chat.name}")
+        group_chat.remove(public_key)
+
+def main():
+    launch_docker_container()
+    load_dotenv()
+    account = Account(backup_folder=os.path.dirname(__file__))
+    account.login(
+        password=os.environ["PASSWORD"],
+        name=os.environ["NAME"],
+        mnemonic=os.environ["MNEMONIC"]
+    )
+    group_chat = GroupChat(account, os.environ["GROUP_CHAT_ID"])
+    warnings = {}
+
+    device = "cuda" if torch.cuda.is_available() else "cpu"
+    account.logger.info(f"Loading Detoxify [{device}]")
+    model = Detoxify("original", device=device)
+    account.logger.info(f"Listening Group Chat {group_chat.name}")
+    for message in account.listen_messages():
+        for chat in message["event"]["chats"]:
+            if chat["id"] != group_chat.id:
+                continue
+
+            threading.Thread(
+                target=check_message,
+                args=(account, chat["lastMessage"], warnings, group_chat, model),
+                daemon=True
+            ).start()
+
+if __name__ == "__main__":
+    main()
diff --git a/pyproject.toml b/pyproject.toml
index 361afc2..9584874 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -36,6 +36,10 @@ agents = [
     "langchain-groq",
     "python-dotenv",
 ]
+group-chat-moderator = [
+    "detoxify",
+    "python-dotenv"
+]
 
 [project.urls]
 Source = "https://github.com/status-im/status-python-sdk"