Klatsch 🐾

Always-on local agent for the OpenClaw ecosystem.

Klatsch runs on each of your machines (Windows or Linux/Pi) and gives OpenClaw eyes and ears on that host:

• 🎙️ Voice assistant — wake-word detection, Whisper STT, edge-TTS, German/English
• 🔔 Notification hub — receive push messages from the gateway and speak them aloud
• 📡 Peer coordinator — Follow-Me routing: whoever is closest to the mic answers
• 🖥️ System awareness — processes, CPU/RAM, disk health, screenshot, clipboard
• 💾 Inventory proxy — exposes local disks and Git repos via HTTP
• ⏰ Reminders — schedule spoken reminders by voice
• 📢 Intercom + broadcast — speak to one peer or all peers at once
• 🌅 Morning briefing — daily 06–09 h gateway briefing (date, weather, agenda)
• 🤖 Ollama integration — "frag Ollama …" routes to a local LLM

The name is German for both "clap" and "gossip" — because when it works, you clap, and it always knows what's going on. 👏

---

Requirements

• Python 3.11+
• FFmpeg (for edge-tts audio decode): winget install ffmpeg
• An OpenClaw gateway running somewhere on the network

---

Installation

# Clone
git clone https://github.com/donapart/klatsch.git
cd klatsch

# Create venv + install deps
python -m venv .venv
.venv\Scripts\activate
pip install -r requirements.txt

Or run the one-shot setup script (Windows, auto-creates a startup shortcut):

powershell -ExecutionPolicy Bypass -File setup.ps1

---

Configuration

All settings are via environment variables (or a .env file):

Variable	Default	Description
GATEWAY_URL	http://localhost:18789	OpenClaw gateway base URL
GATEWAY_TOKEN	opensesame	Gateway auth token
AGENT_ID	main	Agent ID to route voice commands to
HOST_NAME	machine hostname	Displayed in tray and logs
WAKE_WORDS	hey klatsch,klatsch	Comma-separated wake words
TTS_VOICE	de-DE-ConradNeural	edge-tts voice name
WHISPER_MODEL	base	faster-whisper model size (tiny/base/small)
MIC_THRESHOLD	0.015	Voice activity threshold 0.0–1.0
INPUT_DEVICE	system default	Audio input device index
OUTPUT_DEVICE	system default	Audio output device index
VOLUME	100	TTS playback volume 0–100
PEERS	(empty)	Space-separated peer URLs, e.g. http://other-host:7790
SPEAKER_SCORE	0.5	Follow-Me speaker proximity score 0.0–1.0

---

Running

# Normal mode (console)
python klatsch.py

# Background with system tray icon
python klatsch.py --tray

# Discover audio devices
python klatsch.py --list-devices

# Test microphone (5 s recording)
python klatsch.py --test-mic

# Specific devices
python klatsch.py --input-device 3 --output-device 5

---

HTTP API (peer server, default port 7790)

Each Klatsch instance runs a small HTTP server so peers and external services can interact with it.

GET endpoints

Endpoint	Description
GET /health	Status, hostname, presence
GET /inventory	Local disks + Git project list
GET /screenshot	Screenshot as base64 PNG
GET /clipboard	Read clipboard text
GET /processes	Top-20 processes by CPU
GET /syshealth	CPU%, RAM%, disk usage
GET /find-file?q=query	Search files by name

POST endpoints

POST /speak        {"text": "Hello!"}
POST /notify       {"text": "...", "from": "Bot"}
POST /intercom     {"text": "...", "from": "HOST-A"}
POST /clipboard    {"text": "text to set"}
POST /open-app     {"app": "chrome"}
POST /remind       {"text": "Stand-up!", "minutes": 15}
POST /broadcast    {"text": "...", "endpoint": "/notify"}
POST /wake-claim   (internal Follow-Me protocol)

---

Voice commands (German)

Say	Action
"Hey Klatsch, …"	Wake + command
"pause"	Pause current TTS
"weiter"	Resume TTS
"Klatsch, sag host-b: …"	Intercom to peer
"sag allen dass …"	Broadcast to all peers
"öffne Chrome"	Open application
"fokussiere Discord"	Bring window to foreground
"Zwischenablage"	Read clipboard aloud
"erinnere mich in 10 Minuten an …"	Set reminder
"Systemstatus"	Speak CPU + RAM
"frag Ollama …"	Query local LLM

---

Windows Explorer Integration

Right-click any file or use "Send To" to interact with Klatsch directly from Explorer.

Install

powershell -ExecutionPolicy Bypass -File install-context-menu.ps1

This adds three right-click entries and a "Send To" shortcut:

Menu entry	Action
Mit Klatsch vorlesen 🔊	Read the file aloud via TTS
An Klatsch senden 🐾	Ask the AI agent about the file
Mit Klatsch zusammenfassen 📝	Get an AI summary
Send To → Klatsch	Same as "An Klatsch senden"

Uninstall

powershell -File install-context-menu.ps1 -Uninstall

Command-line usage

python klatsch-send.py speak report.txt       # read aloud
python klatsch-send.py ask config.yaml         # ask agent
python klatsch-send.py summarize meeting.md    # summarize
python klatsch-send.py notes.txt               # default: ask

---

Multi-device Follow-Me

Set PEERS=http://other-host:7790 on every host. When the wake word fires on multiple devices simultaneously, the one closest to the speaker (highest mic amplitude) wins — the others stay silent.

---

Optional dependencies

Package	Feature
openwakeword	Lightweight wake-word detection (falls back to Whisper)
pystray + Pillow	System tray icon (--tray mode)
psutil	GET /processes, GET /syshealth, presence watcher, disk watcher
pyperclip	Clipboard read/write

---

License

MIT — see LICENSE.

Built on top of OpenClaw.