SelenaCore Update: Smarter Intent Routing and Optimized Local AI Performance

We have introduced a significant upgrade to how SelenaCore understands and processes your requests to make the system more flexible and responsive. By replacing the traditional LLM classifier with a new freeform assistant tier in commit fed11ab, we've enabled the AI to handle natural language more fluidly without being restricted by rigid categories. To ensure stability on resource-constrained hardware like the Raspberry Pi, we also limited Ollama to load only one model at a time (e5e4eb0), preventing system crashes during heavy tasks. Our core logic is now more precise thanks to improvements in command extraction and better handling of word-to-number conversions (6c9928d). We have also updated our documentation with a comprehensive benchmark matrix for English and Ukrainian, demonstrating the high accuracy of our new embedding classifier (d750e93). These changes mean your smart home is now faster, more reliable, and significantly better at understanding the context of your commands.

What changed:

• Switched to a freeform assistant tier for more natural AI interactions.
• Optimized memory usage by limiting concurrent Ollama model loading.
• Enhanced command extraction and numerical data processing for higher accuracy.
• Published full performance benchmarks for English and Ukrainian languages.