Hi everyone, I’m building TigrimOSR , a Rust-native multi-agent AI workspace. The core idea is Open Loop Engineering : instead of using a fixed hidden agent loop, users should be able to create, edit, inspect, and control their own custom loop. In TigrimOSR, the agent loop is not locked inside the code. You can define it as a YAML profile : which tools the agent can use which MCP servers are available which skills are loaded which model/provider to use custom system prompts ...