dsh-synomega

In AI-assisted chemistry, the most fragile step is rarely the model itself but the act of handing a structure to it. A SMILES string with one substituent shifted by a ring position is an entirely different compound, yet retrosynthesis, route planning, and synthesizability scoring all demand constant switching between text and structure. dsh-synomega embeds five SynOmega model tools directly into the DeepSeek Harness conversation: ask how to make X and receive a step-scored route tree down to purchasable building blocks; ask what a reactant pool can form and get a depth-organized evolution network. After a one-time local download everything runs on your machine with no API key and no remote inference.

The plugin ships with a Ketcher structure editor so you can draw a molecule into the conversation instead of typing SMILES by hand, eliminating transcription risk. In the browser, results render as cards: each candidate reaction shows reactants to product with its score, routes appear as trees with per-step scores on the edges and unpurchasable starting materials highlighted. All rendering is done by locally bundled smiles-drawer and d3-hierarchy, works fully offline, and reopening an old conversation reproduces the same picture.

Best suited for researchers and educators in medicinal synthesis, natural-product total synthesis, or reaction-condition optimization who need a local, reproducible, network-independent setup. All inference stays on a single machine, results are reproducible, and no structures leave the host.