When you operate multiple AI agents and automated tasks on the same server — data collection agents, summarization agents, API LLM calling agents which incur costs — it is very difficult to grasp the overall picture: what is running, for what purpose, and how much it costs. Quartermaster was created to address exactly this issue.

Quartermaster is a read-only operational assistant. You point it to the server, and it tells you what each agent is, what it is doing, how much API/LLM cost is incurred, which components it depends on, and what will be affected if it stops. Each conclusion is accompanied by evidence and a level of confidence; when there is insufficient basis, the system clearly states "undetermined" instead of guessing.

The core difference: an AI monitoring tool but does not use AI itself. Quartermaster operates fully deterministically — based on rules, dependency graphs, and actual metrics, without involving LLM in the monitoring process. This way, it does not generate false alerts due to "hallucinations," cannot be prompt-injected through the logs it reads, and does not incur additional token costs.

Key features:

• Scans repositories, processes, services, containers, and ports — completely read-only.
• Constructs dependency diagrams and detects configuration drift over time.
• Cost analysis: aggregates by provider, allocates costs to the correct agent, alerts when exceeding budget or unusual spending, and traces unidentified ownerships.
• Incident reports with recommendations on "what to check first".
• Only observes and recommends, never intervenes on its own. Supports sending alerts via Telegram (optional).

Quartermaster is an open-source project (MIT license), prioritizing Linux platforms. I am currently looking for beta users and welcome feedback from the community.

GitHub: https://github.com/Mikeung/quartermaster