Benchmarks ยท aiDragon
aiDragon Benchmarks
Why aiDragon is unlikely to carry the same local performance benchmarks as the other Dragons.
An agent Dragon, not a local model host
aiDragon is the agent-facing Dragon: workspaces, MCP surfaces, memory and responsible access. It is about orchestrating and safely exposing tooling, not about running inference on the box itself.
For that reason this page is unlikely to be populated with the kind of throughput or latency figures seen on the other Dragons. The work it does is agent- and integration-shaped, which does not reduce to a single reproducible local benchmark in the same way a web, mail, DNS, tunnel or database service does.
Why no local model benchmark
Saphira Linux uses musl libc. We do not run local models, and the NVIDIA-Open driver support needed for local GPU inference is not available under musl. Without local model execution there is no local inference workload to benchmark, so aiDragon will mostly remain a methodology note rather than a results page.
If a meaningful, reproducible agent-side measurement appears later, for example around tool-call routing or memory operations, it would be recorded here with the same discipline as the other benchmark pages. Until then, aiDragon is the exception that proves the rule: benchmark what you can actually run and reproduce.