tqdaemon

English | Polski | 中文 | العربية

tqdaemon runs long jobs on your own machines without melting them. Describe each job in YAML. A local daemon starts work only when there is enough CPU, RAM, and GPU memory. Watch progress in a web UI, the tq CLI, or from an AI coding agent (MCP).

Not Celery, Kubernetes, or CI. Not the 2011 AMQP/Debian rojkov/taskqueue (simpletaskqueue).

Why use it

• Resource-aware admission (CPU / RAM / GPU VRAM) on one PC or with remote workers
• YAML jobs with checkpoints, deps, priorities, and stable error codes
• Built for agents: submit, wait, diagnose over MCP
• Local-first — no cloud SaaS account

Best start: ask your agent

Tell your coding agent (Cursor, Claude Code, …) to use tqdaemon for long jobs: wire MCP (tq mcp), submit YAML tasks, and wait_for_task instead of blocking the chat. Setup: docs/mcp.md.

Install

Need: Python 3.12+, Git, uv.

From PyPI: uvx --from tqdaemon tq doctor (or pipx install tqdaemon). From source:

git clone https://github.com/wwylon/tqdaemon.git
cd tqdaemon
uv sync
uv run tq doctor
uv run tq start

Open http://127.0.0.1:8765. Try the sample in examples/minimal (register it as a project, then tq sync / enable / tq wait — see that folder’s README).

Optional Windows toasts: uv sync --extra toast. Autostart: scripts/install_tq_autostart.ps1.

Docs

Advanced: Security & privacy · Task standards · Scheduler · Workers

Transparency

We do not phone home. Public signals are GitHub stars (badge above) and, after a PyPI release under a distinct package name, download counts you can verify yourself. Maintainers may use GitHub Traffic Insights (clones ≠ installs). Details: docs/security.md.

License

MIT — see LICENSE · Changelog