I’m building a local-first multi-model AI agent stack on macOS.

I’m building a local-first multi-model AI agent stack on macOS. It connects cloud and local LLMs, MCP tools, browser/desktop automation, task routing, and self-healing recovery. One of my main goals is to make long-running agents survive real-world failures such as model limits, network interruptions, tool errors, and context growth without needing constant manual intervention. I’m very interested in testing where Jev could fit into this stack, especially around reliable tool use and autonomous multi-step workflows.