Genome Computer offers private AI chat for exploring personal genome data

Why it matters

Ball is turning genome privacy into infrastructure: a consumer chat proves the model, while Genome Computer's API can carry the same data layer into other health products.

David Ball (@davieball) is offering Genome Intelligence, a conversational interface for exploring genomic data with general-purpose AI models while keeping the user's complete genome in a separate store.

A repost from Robert Scoble's account carries an announcement attributed to Ball, but the original post's date is not verifiable from the available source.

Genome Intelligence announcement on X

Genome Intelligence is the consumer-facing layer of Genome Computer, the San Francisco company Ball has been building around a larger thesis: genomic data should be portable, legible to AI systems, and controlled by the individual whose DNA produced it.

That thesis is a sharp turn for a founder who entered startups through marketplaces. An Australian startup profile published in 2009 identified Ball, then 22, as founder and managing director of Zebra International. He had launched Askatradie.com.au, a service connecting consumers with qualified tradespeople. The profile said Askatradie attracted 4,000 customers in two weeks, a figure attributed to the publication. An earlier Ball project, a GPS-based mobile content security concept called Zspot, had been shortlisted in the University of Queensland Business Enterprize competition.

Seventeen years later, Ball is applying that early startup experience to the trust problem surrounding consumer genetics. In a July 1 post indexed on his public founder profile, Ball wrote that Genome Computer was started to "right the wrongs" that caused people to lose trust in the category. Genome Computer's answer is a combination of technical separation, portable files, and corporate restrictions on the reuse of individual genetic data.

The model gets a slice, not the whole genome

Genome Computer says Genome Intelligence stores a user's full genome separately from the AI model handling a question. For each query, the service selects the relevant variants and sends those without information that identifies the individual.

A user asking about caffeine metabolism, for example, would expose the model to the portion of the genome needed to answer that question rather than uploading millions of variants into a chat session. Users can also add blood results, diagnoses, and medical records, according to Genome Computer's product page.

Genome Computer says hosted genomes are re-annotated monthly as research and reference databases change. That recurring interpretation layer is central to the subscription model: DNA remains largely fixed, while scientific conclusions about particular variants can be revised.

Genome Intelligence costs $15 per month for customers with a Genome Computer file. Genome Computer currently lists whole-genome sequencing at $299 for 3x coverage, $599 for 30x, and $1,499 for 100x. Converting an existing VCF or consumer DNA file into Genome Computer's format costs $99. Genome Computer says it is sequencing and annotating 1,000 genomes, though that volume has not been independently verified.

The service carries an important boundary. Genome Computer says its files are raw data rather than medical tests and directs users to physicians or clinical genetics services for clinical interpretation. Genome Computer also warns that its lower-cost 3x sequencing option cannot call rare variants with clinical confidence.

Ball built the file format before the chat

Genome Intelligence sits on work Ball released months earlier. In April 2026, Genome Computer published .genome/1.0, an Apache-licensed specification designed to make consumer genome files readable by AI agents.

Traditional VCF files were designed for bioinformatics pipelines whose operators already understood the surrounding schemas, headers, annotation conventions, and reference databases. Genome Computer's format separates variants, interpretations, evidence, and decision rules into typed, versioned, queryable structures. The original VCF ships alongside the new bundle rather than being discarded.

According to the open-source repository's README, the project includes the specification, a JSON schema, a Python reference reader and validator, a synthetic example bundle, benchmarks, and reference VCF conversion scripts. The README cautions that the converter is a reference implementation rather than Genome Computer's production pipeline. The repository also includes readmygenome.md, a set of instructions intended to help Claude, Codex, and other agents query the bundle without improvising missing context.

Genome Computer claims the format uses three to 10 times fewer tokens than annotated VCF files and reduces factual errors by 10 to 20 times. The repository contains reproducible benchmarks, but Genome Computer has not identified an independent evaluation supporting those headline ranges. The repository's README calls v1.0 an unfinished standard and lists structural variants, phasing serialization, and multi-sample bundles among its open questions, with structural variants and multi-sample bundles deferred to later versions.

The consumer chat is one part of a larger platform

Ball's broader play becomes clearer in Genome Computer's developer documentation. Genome Computer is offering an API for teams building nutrition, longevity, recovery, and clinical-adjacent products. The API covers sequencing orders, hosted genome storage, interpretation artifacts, prebuilt insight panels, bounded genome questions, and webhooks for updated interpretations.

That makes Genome Intelligence a reference product for the underlying runtime. Ball can use the consumer interface to demonstrate the privacy architecture and monthly re-annotation system, while other software companies build their own interfaces on Genome Computer's API.

Other genomics providers are moving toward recurring AI interpretation. Human Longevity currently sells a $599 whole-genome product with an AI application and annual re-analysis. Genome Computer is competing through portability, monthly updates, model choice, and its claim that a user's complete genome does not need to enter each model conversation.

Privacy is therefore part of the product design and the sales proposition. Genome Computer is operated by Genetic Superintelligence Company, a Delaware Public Benefit Corporation. In a June statement about that structure, Genome Computer said it would not sell or license individual genetic data, share it with researchers or drugmakers, or use identifiable genomes to train AI models.

Genome Computer also says private orders can be processed using a Genome ID and access code without a permanent customer email attached to the genome.

Those commitments remain company assertions. Public-benefit status creates a governance obligation, but it does not independently validate Genome Computer's security controls, identity separation, model-routing system, or deletion process. Ball is asking customers to trust both an architecture and an institution with data that cannot be reset after a breach.

His bet is that consumer genetics can earn that trust by giving people a useful file, a choice of AI models, and fewer reasons to surrender the underlying asset.