We introduce ARCHE, an autonomous agentic system that couples a general‑purpose reasoning model, a domain‑specific computational chemistry model, and a structured tool registry. The system first parses a scientific query, then generates and ranks mechanistic hypotheses, orchestrates computational workflows, and iteratively refines conclusions within a closed loop based on computed evidence.
ARCHE’s capabilities are validated across three increasingly demanding scenarios: (1) reconstructing the stereocontrolling transition state and confirming the mechanism of a previously reported asymmetric catalytic reaction; (2) proposing and validating a plausible radical pathway for a newly discovered α‑iodoboronate C‑I cleavage reaction through iterative hypothesis refinement; (3) identifying an interpretable descriptor that governs selectivity in nickel‑catalyzed migratory cross‑coupling reactions.
By coupling agentic reasoning with rigorous computational validation, ARCHE advances autonomous mechanistic discovery and establishes a foundation for broader machine‑assisted chemical research. The code is publicly available at https://github.com/JetAstra/Arche-Harness
Review