Iktos

Generative AI is the product. Makya, sold as a SaaS platform, generates novel molecules under a target product profile with synthesis built into generation, having been trained on millions of reactions; Spaya is an AI retrosynthesis platform that finds routes to commercially available starting materials. Iktos also runs its own robotics for synthesis and testing and an in house pipeline, both fed by those models. Remove the models and the software products cease to exist.

The oversight structure is described at the design step and the automation boundary is not. In Makya the chemist defines the target product profile, reviews generated compounds against every score and selects which to make; Spaya proposes routes for review. Downstream, Iktos Robotics automates ordering, synthesis, purification, analysis and testing, and the company describes its direction as "fully autonomous labs". What is not published is which decisions the automated lab takes without a person, such as which compounds advance to the next cycle, or the thresholds that stop a run.

The approach and training data are described, with releases named but models not identified. Makya is described as generative AI trained on millions of reactions, exploring a chemical space of about 10^27 molecules with synthesis constraints built into generation; the January 2026 release introduced two new generative models that design molecules and their 3D poses from a known active ligand without docking, and Makya 2.0 redesigned the analysis interface.

Release factsheets exist (version 1.18), and the publications list covers methods such as flow matching pose generation. The models' architectures and versions in the running product are not identified on the product pages.

A structurally short chain: the models are Iktos's own, and customers can keep the software in their own environment. Makya and Spaya are Iktos's generative and retrosynthesis models, trained on reaction data, with no outside model provider described; customers can connect their own external models through an API. For the on premises and private cloud options, a customer's chemistry need not leave its environment. No subprocessor list or hosting provider is published for the SaaS option, and Spaya draws on outside data providers that are not named.

Published research and preclinical programs with data, short of independent validation. The resources page lists the company's scientific publications and posters, including a 2026 paper on combining AI with human expertise to find PolQ inhibitors and work on compound selection guided by cell painting.

The in house pipeline shows programs made with Makya and Spaya: an MTHFD2 inhibitor lead series for inflammatory and autoimmune disease (IKT525, IC50 below 5 nM and more than 1,000-fold selective over MTHFD1, company data, patent WO2026132447) and an amylin receptor program in obesity. The company names many pharma collaborators and states discovery to a preclinical candidate in under two years, without method. No candidate is in a clinical study.

The live question is protection of customers' chemistry, and the published answer is architectural. Makya is offered as SaaS and, per earlier company material, on customer premises or inside a customer's own private cloud, which keeps a customer's compounds and targets out of a shared environment. No specific commitment on training use or retention is published, though the one pager describing those terms may set one out; the website privacy statement covers website visitors only. No patient data is in the workflow. Safety engineering is not addressed.

No position is stated and no privacy document reaches the product. The only privacy document published is the website privacy statement, which covers people who contact the company through its site. Makya and Spaya work on chemical structures and assay data, and their customers are pharmaceutical and biotech organizations rather than covered entities, so protected health information is unlikely to be involved. The company does not say so, and there is no service privacy notice.

Nothing independent is published. The site shows no security page, trust center or certification, and the website privacy statement covers website data only. The Makya one pager may cover security terms. For a French company licensing design software to European and Asian pharmaceutical groups, contractual data processing terms will exist but are not published.

No device claim is made and the products are scoped as chemistry design software. Makya and Spaya design molecules and plan syntheses and make no diagnostic or treatment recommendation, so no clearance applies. Iktos's own programs are preclinical; regulatory standing for molecules designed with its tools sits with whoever develops them.

Method research is published; governance of the models is not. Iktos lists scientific publications and posters on its resources page, from synthesis planning (2023) to 3D pose generation by flow matching and a 2026 paper on combining AI with human expertise to find PolQ inhibitors.

No governance framework, model card, evaluation of where Makya's generation or scoring performs poorly, or monitoring of model changes is published, and no paper on those questions is named on the pages that sell the product. Safety engineering is not addressed.

Mechanisms let a chemist check an output, with nothing standing behind it. Makya shows its scores, SAR and ADMET predictions and 3D interactions for each generated compound, with filters for choosing among them, and Spaya displays the retrosynthetic routes behind a proposal, so the customer's chemists review and select before anything is made. No accuracy or hit rate for generated compounds, warranty, remediation or fee consequence for designs that fail is published.

No clinical record integration exists: Iktos sells design software and services to research teams, and no clinical record system is involved. For a pharma or biotech buyer the record systems that matter are its research systems: electronic lab notebooks, laboratory information systems or compound registries, research data platforms and cloud data environments.

The integration Iktos describes is generic: Makya accepts a project's own data and connects external models and scoring functions through an API, can be driven as a Python package, and Iktos has announced engine APIs for agentic workflows and computational chemistry pipelines; Spaya is said to integrate with various data providers.

The company names no electronic lab notebook, compound registry, research data platform or customer deployment, and does not say which customer systems run connected in production or in which direction data flows.

Deployment options are stated, with residency and the processing path left open. The Makya page describes a SaaS platform; company material also offers installation on the customer's premises or in the customer's own private cloud, with access through the interface or a Python package, which puts a customer's chemistry under its own control on those options. The one pager with those terms may spell them out in more detail. For the SaaS option no hosting provider, region or tenant separation is published (the website itself is hosted by Vercel).

No price is published, and how the products are sold is discoverable. Makya and Spaya are licensable software, with a demo request as the route in, and Iktos also runs discovery collaborations and its own pipeline; a strategic collaboration with Servier was reported at a potential value above €1 billion, which describes one negotiated deal rather than a price. No rate card, unit of charge or tier is published for the software, though a licensed SaaS product could publish one.

Coverage is named, with validation behind part of it. The products and pipeline cover small molecule design and synthesis planning, with robotics biology for protein protein and RNA protein interactions, molecular glues and PROTACs, and in house programs in inflammation and autoimmune disease (MTHFD2) and obesity (amylin receptor). Validation is strongest in the in house programs and published work; collaborations span many named pharmaceutical partners without published results by area.