Six claims. Six sources.
Every async def agent-callable function enumerated at build time from the canonical Python source.
View full catalogReAct agent definitions across all crew folders, generated from the platform crew definitions.
View all personasCloud, local, and open-router providers. Add one entry, the picker and codegen pick it up.
View providersRust engine, in-process monotonic clock. 89,033 samples, engine 0.4.48. Full pipeline p50: 14.4 µs.
Full latency matrixPython agentic runner: tool dispatch, RAG retrieval, model-call overhead, HITL gate, pipeline orchestration.
Full framework benchEverything on this page is reproducible from a fresh clone. Bench harnesses in the public OSS repo.
OSS repoHow do I verify Melaya's benchmark claims?
Melaya publishes 7279 tools, 111 subagents, and 29 AI providers in build-time-generated JSON catalogs. The Rust engine headline (310 ns p50, n=89,033) is measured in-process with a monotonic clock and reproducible from a fresh clone via cargo bench. The agentic framework suite covers 16 measured metrics, reproducible via pytest benches/. Source links are on each sub-page.
