Watch any process. Get notified on Telegram. Control it remotely.

---

qara is a lightweight daemon that monitors processes and sends real-time notifications to Telegram. Start a long-running job, close your laptop, and stay in control from your phone.

Built for ML practitioners who run multi-hour training jobs, but works with any process.

qara run python train.py --name "gpt-finetune"
# => Telegram: "Process 'gpt-finetune' started (PID 41592)"
# => ... hours later ...
# => Telegram: "Process 'gpt-finetune' finished (exit 0, 3h 42m)"

Features

• Spawn or attach — start a new process with qara run or monitor an existing one with qara attach <pid>
• Telegram notifications — get notified on start, finish, and crash with configurable stdout tail
• Remote control — send /status, /kill, /history, /logs from Telegram
• Daemon mode — runs as a user-level service (systemd, launchd) — no root required
• JSON output — --format json on status and history for scripting

Quickstart

Install

# Install globally (recommended)
pipx install qara

# Or with uv
uv pip install qara

Configure

qara config init

This creates ~/.config/qara/config.toml. Open it and add your Telegram bot token and user ID:

[telegram]
bot_token = "123456:ABC-DEF..."
allowed_user_ids = [your_telegram_user_id]

How to get a bot token and user ID

1. Message @BotFather on Telegram and send /newbot
2. Follow the prompts to name your bot — you'll receive a token like 123456:ABC-DEF...
3. To find your user ID, message @userinfobot and it will reply with your numeric ID
4. Important: Send any message to your new bot first so it can message you back

Start the daemon

qara daemon start

Run a process

qara run python train.py --name "experiment-1"

That's it. You'll receive Telegram notifications when the process starts, finishes, or crashes.

Usage

Process management

# Run a process
qara run python train.py --name "my-job"

# Attach to an existing process
qara attach 12345 --name "background-job"

# List watched processes
qara status
qara status --format json

# View completed runs
qara history --last 10
qara history --format json

Telegram commands

Once the daemon is running, send these commands to your bot:

Command	Description
/status	List all watched processes
/kill <name>	Send SIGTERM to a process (escalates to SIGKILL)
/history	Show recent completed runs
/logs <name>	Get last N lines of stdout

Daemon management

qara daemon start              # Start in background
qara daemon start --foreground  # Start in foreground (for systemd/launchd)
qara daemon stop               # Stop the daemon
qara daemon status             # Check if daemon is running

Install as a system service

# Auto-detects systemd (Linux) or launchd (macOS)
qara install

# Preview what would be installed
qara install --dry-run

# Remove the service
qara uninstall

Configuration

Full config.toml reference:

[daemon]
log_level = "INFO"  # DEBUG, INFO, WARNING, ERROR

[telegram]
bot_token = "YOUR_BOT_TOKEN"
allowed_user_ids = [123456789]

[telegram.notifications]
on_start = true
on_finish = true
on_crash = true
stdout_tail_lines = 20  # lines of stdout included in finish notification

[commands]
enabled = ["status", "kill", "restart"]
kill_timeout_seconds = 10

[commands.allowed_scripts]
# alias = "/absolute/path/to/script.py"

Architecture

qara daemon start
      │
      ▼
┌─────────────┐     ┌──────────────┐     ┌──────────────┐
│   Watcher    │────▶│  EventEngine │────▶│  Telegram    │
│  (per proc)  │     │  (pub/sub)   │     │  Channel     │
└─────────────┘     └──────┬───────┘     └──────────────┘
                           │
                    ┌──────┴───────┐
                    │   Plugins    │
                    │ (GPU, loss)  │
                    └──────────────┘
      ▲
      │ IPC (Unix socket)
      │
  qara run / qara status / ...

• Watcher spawns or attaches to processes via asyncio.create_subprocess_exec
• EventEngine is sequential async pub/sub — handlers receive events in subscription order
• Plugins subscribe directly to the engine and receive all events including stdout/stderr lines
• NotificationBus filters internal events before routing to channels
• IPC uses newline-delimited JSON over Unix sockets (Linux/macOS)

Development

git clone https://github.com/warptengood/qara
cd qara
uv sync --group dev

# Run checks
uv run ruff check src/       # lint
uv run ruff format src/      # format
uv run mypy src/             # type check
uv run pytest                # tests

See CONTRIBUTING.md for guidelines.

License

MIT — see LICENSE for details.