Scenario

Designing a metal complex for a target property — color, spin state, stability, or catalytic behaviour — means choosing the metal and ligand set well. Predicting coordination-complex properties up front triages candidates before the glovebox.

Worked example. A luminescent complex needs a ligand set that gives the desired geometry and stability.

Agent workflow

1. Read the candidate metal centers and ligand sets.
2. Coordination complex properties — predict geometry, coordination environment, and the relevant complex properties.
3. Rank candidate complexes against the design target.

Demo output

A ranked set of candidate complexes with predicted geometry and target properties, plus the ligand feature driving each result. The magic moment: an unexpected ligand combination gives the target geometry and stability, worth prioritizing for synthesis.

Deterministic vs LLM

• Deterministic — the coordination-complex property predictions.
• LLM — the ligand-selection narrative.

Every property is a validated calculation. The agent reads and routes; the engine decides.

Limits

Screening-level design support, not a substitute for synthesis and characterization. It focuses the ligand search; the lab confirms the complex.