What is spark-signals?

spark-signals is a monitoring tool for a specific piece of NVIDIA hardware called the DGX Spark, a machine designed for running AI workloads. It watches over the health of the Linux operating system, the NVIDIA graphics hardware, certain configured background services, and any AI model servers running on the machine, then packages that information into structured messages. The project is made up of two parts. The first, called spark-agent, collects measurements like CPU usage, memory, network activity, disk activity, and temperature, and publishes them to a lightweight messaging system called NATS. It can also simply print the same information to the screen for someone checking on a machine directly. The second part, called spark-otel-bridge, listens to those published measurements and converts them into a format called OTLP, a common standard used by monitoring and observability platforms, so the data can flow into dashboards and alerting tools that already speak that standard. A design choice the author highlights is that when a measurement cannot be collected, the tool marks it as explicitly missing rather than quietly reporting a fake zero, which the author says avoids misleading anyone looking at the data. The project also supports optional adapters for several AI model serving tools, including SGLang, vLLM, llama.cpp, and any service that speaks the OpenAI compatible API format. According to the project's own roadmap, this is still an early stage prototype: the core data collection and publishing pieces exist, but the visual dashboard and some operational hardening work have not been started yet. The README includes a full user guide and a set of installer scripts meant for setting this up as a proper system service in a production environment, separate from any individual user's login account, along with scripts for validating that the installation is working correctly. The project is written in Rust and is released under the MIT license.