Helixstreet builds the Sovereign Twin — a private, in-cloud AI that reads your whole genome, talks to you about it, and helps you act on it. Your data never leaves a perimeter you control.
Built in Basel since 2017 by a team that has spent years on long-read sequencing and deep learning for DNA, a member of the NVIDIA Inception program since 2019, renewed again for 2026, and a participant in the Google AI Agents Challenge. The twin runs on Vertex AI and Gemini 2.5 Pro inside a Google Cloud project you own.
Gemini's million-token context lets the twin read across your full sequence — variants, regions, and structure — in one go.
Runs inside your own Google Cloud project on Vertex AI. No third-party training, no shared inference, no data egress.
Ask in plain language. The twin grounds every answer in your sequence and the literature, and shows its work.
Upload a VCF or FASTA into a project you own. Encrypted at rest, scoped IAM, no copies anywhere else.
Vertex AI provisions Gemini 2.5 Pro inside your perimeter and indexes your sequence into long-context working memory.
Pharmacogenomics, carrier status, ancestry, variant interpretation — grounded in your data with citations you can audit.
Short answers. Email hello@helixstreet.ai for anything longer.
No. The Sovereign Twin is an interpretation and research tool, not a medical device. It helps you read and understand your sequence, with citations you can check. Decisions about diagnosis, treatment, or medication belong with a qualified clinician or genetic counselor.
A VCF (variant calls) or FASTA (raw sequence) from any provider: clinical whole-genome sequencing, research pipelines, or consumer services that let you export raw data. Larger files simply take longer to index.
In a Google Cloud project that you own. Helixstreet deploys into your perimeter and runs inference on Vertex AI there. We never hold a copy, and nothing egresses to a shared service. Deleting the project deletes everything.
No. Vertex AI does not use customer prompts or data to train foundation models, and Helixstreet has no access to your sequence or conversations in the first place.
During the private preview, Helixstreet is free. You pay Google Cloud directly for the storage and Vertex AI inference your project consumes, at your own negotiated rates. There is no markup and no separate Helixstreet bill.
A Basel-based team working on AI for DNA since 2017, with a second base in San Francisco. Helixstreet has been a member of the NVIDIA Inception program for AI startups since 2019, and the membership was renewed again for 2026. We are also a participant in the Google AI Agents Challenge. The Sovereign Twin is the current product of that work.
No. The app walks you through connecting a project and grants itself only the scoped IAM roles it needs. If your organization has its own security requirements, CMEK, VPC Service Controls, and audit logging are all yours to configure.
Helixstreet deploys into your GCP organization. CMEK, VPC-SC, and audit logs are yours to configure.
Vertex AI does not use your prompts or sequences to train foundation models. We don't either.
Every answer traces back to a region of your sequence and a source it cites. No black boxes, no magic.
Who should hold the most personal dataset a person will ever produce? We think the answer is obvious.
Helixstreet started in 2017 at the Technologiepark in Basel, working on long-read sequencing and deep learning to find the structural variants that reference-based pipelines miss. In 2019 we joined the NVIDIA Inception program, NVIDIA's accelerator for AI startups, which backed that work with compute credits, GPU access, and engineering support. The membership has been renewed every year since and remains active in 2026. Helixstreet is also a participant in the Google AI Agents Challenge, where the Sovereign Twin is built on the same Vertex AI and Gemini stack it ships on.
The Sovereign Twin is where that research landed: a conversational model over your whole genome that lives in your own cloud perimeter. Same team, same question, better tools. Today we work from Basel and San Francisco.
Open the app, connect a project, and try a conversation. Private preview — no waitlist for the curious.