Curious about performance: do you have benchmarks on latency/throughput compared to pure‑Python agent frameworks? The Rust core seems ideal for low‑latency tool calls.