megasoft1978/strim-engine is a Swift-based local inference engine that runs Qwen-family models on Apple silicon Macs. The repository description lists Qwen3.8-27B and MLX as its targets, and it also presents a sandbox harness for coding agents. The latest release shown on GitHub is v0.1.7. However, the provided material does not list what was newly added or changed in this version, so the specific changes in v0.1.7 cannot be confirmed.
This project focuses on running models locally and providing an API that coding agents can connect to. According to the README summary, it gives a local API address and a server run example, so you can look at how an agent uses the server. The server is introduced as working in OpenAI format, and an optional menu bar app is also described. It does not stop at explaining only the model execution engine: the distinctive point is that a single repository covers the information needed for agent connection and server operation. The repository's language is Swift and its license is MIT.
The README is summarized as having published measurement results and measurement environments for concurrent requests and for cases where multiple agents share context. However, the provided FACTS contain no specific figures or environment conditions, so there is no basis for comparing performance or applying it to other environments. The isolation scope of the sandbox harness is also not explained in the given material. It offers a starting point for local agent configuration, but performance and isolation level are things that need separate confirmation.