OnePot

Onepot (Onepot AI, Inc.) is a San Francisco-based company that provides on-demand custom synthesis of small molecules using an automated laboratory paired with AI agents. Customers submit a target structure or select compounds from an enumerated, make-on-demand catalog; Onepot's AI chemist, named Phil, plans the route and conditions, robotic hardware executes the chemistry, and the resulting compounds are shipped to the customer as dry material or as solutions in plates or vials. The company positions synthesis — the "make" step of the design-make-test-analyze cycle — as the principal bottleneck in early-stage drug discovery, and reports average turnaround of about five days versus the six to twelve weeks it describes as typical.

The company's first lab, POT-1, uses liquid handlers, plate sealers, plate-based reaction formats, and a robot in a glovebox, covering reagent supply and storage, reaction setup, workup, purification and QC. Phil is implemented as a tool-using agent that can call lab functions such as sealing plates, write liquid-handling protocols, run code, search literature, interpret LC/MS traces, form hypotheses about byproducts and failures, and design follow-up experiments; the company says every experiment is logged with full process detail, including temperature and reagent additions, to make runs reproducible and to serve as training data. At launch, five reaction classes were available to commercial partners — reductive amination, Buchwald-Hartwig amination, Suzuki-Miyaura coupling, amide coupling and acylations — with more in development.

Onepot also publishes research on reaction prediction with graph neural networks, closed-loop synthesis optimization and foundation models trained on large-scale experimental data. Preprints include onepot CORE, an enumerated chemical space of 3.4 billion molecules supported by seven reactions with validation on success rate, timelines, purity, identity (including NMR confirmation) and assay suitability using a series of DPP4 inhibitors, and onepot-Bench 0, a benchmark suite for whether language models can make the decisions required to execute synthesis. A later release, onepot CORE 2 Triangulum, is described as covering 30 billion compounds across 16 reaction classes.

Founding story

Cofounders Daniil Boiko and Andrei Tyrin first met in a chemistry lab at M. V. Lomonosov Moscow State University and later reconnected in the United States, where Boiko was a PhD candidate in machine learning for chemistry at Carnegie Mellon University (with a bachelor's and master's in organic chemistry) and Tyrin studied computer science at MIT. Boiko observed drug discovery teams skipping promising compounds because they appeared too difficult to synthesize, while Tyrin, working on computational drug-discovery pipelines, saw generative models produce candidate ideas in hours that labs took months to make. They concluded that investment was concentrated in molecular design rather than in the harder problem of making molecules, and set out in March 2024 to build a lab that would perform synthesis automatically.

Business model

Onepot sells synthesized compounds as a service rather than software: customers order from a catalog of make-on-demand molecules or submit their own structures, and the company manufactures and ships the physical material. Its stated alternative-to positioning is against in-house medicinal chemistry teams and overseas contract research organizations.

Revenue comes from customer orders for synthesized compounds delivered as dry compounds or as solutions in plates or vials; the company describes early commercial partners in biotech and pharma using the platform.

Traction

Early commercial partners in biotech and pharma are using the platform and reportedly receive new molecules 6-10x faster, with an average turnaround of about five days. The ordering library covered 1.9 million compounds at launch, many never previously synthesized, and the enumerated onepot CORE 2 Triangulum space is described as 30 billion compounds across 16 reaction classes. The company states Phil ran more reactions in a single month than a typical graduate student would during an entire PhD, and reports improved purity after rolling out its automated purification method construction. Onepot disclosed $13 million in total funding in November 2025.

Latest developments

In November 2025 Onepot exited stealth with $13 million in funding and publicly introduced its automated lab and the Phil AI chemist, stating that proceeds would fund a second, larger San Francisco lab across the street from its headquarters, additional biopharma customers, team growth and its compound discovery engine, with roughly two years of runway. The company subsequently announced onepot CORE 2 Triangulum, an enumerated space of 30 billion compounds across 16 reaction classes, and is hiring across chemistry, software and robotics.