Hi HN!
I made Sitrep, which is an AI incident copilot that sites alongside you during incidents and watches your screen. You can chat with it, and ask questions ("what was that trace I saw?") and it uses vision to answer your questions, providing evidence. It can also integrate with existing observability tools too, I've added Grafana as an example. This means the agent can not only reference context from vision, but also verify context by using actual logs/traces. To avoid duplicate frames, it uses a tile strategy and a hamming distance threshold. It also has blackout mode for when you want to pause it mid incident, and comes with a cost dashboard and settings/model toggle to minimize cost. It's open source, check out the repo and feel free to provide any feedback.