Based on the available codebase memory context, I cannot determine how asyncio
semaphore specifically controls Ollama concurrency in local mode. The retrieved
entities show:

1. `_should_use_cloud` - Determines cloud usage #main.py:986
2. `MEMORY_MODE` configuration with cloud/hybrid modes #config.py:30-45
3. `_background_scan` - Uses Semaphore(4) for tree-sitter and Semaphore(2) for
   LLM summarization during scanning #main.py:1827

None of these entities contain the specific concurrency logic for `_query_ollama`
at runtime. If this routing exists in the codebase, it's not present in the
retrieved memory entries.