GeneMachine is a free educational browser app and an Agent Skill for genomic research. The app helps you understand established gene–medicine evidence, prepare questions for a care team, and inspect a compatible consumer DNA export on your own device.
Built with JavaScript and Cloudflare Workers.
You can use the medicine library without a DNA file, account, subscription, or paid kit. Optional DNA checking runs in a local Web Worker. Genetic data is never sent to an AI service by the browser app.
Open GeneMachine to use the educational app in your browser. No installation is required. For the source-backed product evaluation, read the published evaluation.
To run your own local copy:
Install Node.js 22 or later and Git, then:
git clone https://github.com/baney75/GeneMachine.git
cd GeneMachine
npm startOpen http://127.0.0.1:4173/. The app requires no runtime packages or API keys. Search for clopidogrel or CYP2D6, open a topic, and choose Add this question. Or choose Write a question for any medicine, including one the library does not cover. Edit the wording, add personal notes, and move your main concern to the top. Keep up to five questions, then copy them or download a printable appointment worksheet. Open the downloaded HTML file in your browser to print or save as PDF.
To see the DNA workflow without using anyone’s data, choose Try the synthetic example. You should see OBSERVATION SUPPORTED, an explicit synthetic label, and a single rs4149056 observation with its limits. Clear this DNA now clears the displayed finding.
For a private file, read the privacy explanation, consent to local-only processing, and choose an original .txt, .tsv, or .csv export. Extract ZIP files on your device first. See the official 23andMe and AncestryDNA instructions. No new kit is needed to use the library.
- Seven source-linked learning topics: statins, clopidogrel, thiopurines, fluorouracil/capecitabine, codeine/tramadol, warfarin, and tacrolimus. Search recognizes medicine names, several brand names, genes, and categories.
- General evidence, individual unknowns, and testing questions. Library topics are educational; they are never matched to your DNA or presented as your prescriptions.
- An editable appointment worksheet for any medicine: five questions, personal notes, priority ordering, plain-text copy, and printable HTML. Custom medicines are marked Medicine not assessed; edited questions and notes are labeled as your writing. Curated source links and caveats travel with their topics. An absent topic does not establish that genetics is irrelevant.
- File-quality checks, chromosome row counts, declared genome build and strand, no-calls, duplicate markers, and an original-byte SHA-256 digest.
- One gated SLCO1B1 rs4149056 observation, or a specific reason the tool cannot report it. Build 37/38, forward strand, exact coordinate, valid alleles, and a unique called row must agree.
- A separate source-backed DNA discussion report, with the original filename and full genotype table excluded.
The checker does not call star alleles, diplotypes, phenotypes, CYP2D6, HLA, copy number, or structural variants. A missing marker is unknown. A high file call rate does not establish pharmacogene coverage. The medicine library does not expand the DNA checker’s analytical scope.
Raw DNA is read and parsed in a worker that is terminated after analysis or cancellation. Only aggregate quality checks and the narrow supported finding reach the interface. No analytics, accounts, localStorage, sessionStorage, IndexedDB, or external API is used. Questions, notes, and unsaved drafts stay in this tab's memory until you copy or download them; clearing the list or leaving/reloading the page removes that in-app writing. Withdrawing consent clears the derived DNA result. Reset cannot delete your original file, clipboard contents, or downloaded reports. Copies can contain sensitive findings or personal writing; store and share them deliberately. Browser extensions, a compromised device, or a modified app are outside these software guarantees.
This app provides educational exploration and discussion support. It has no clinical validation or regulatory clearance. It does not diagnose a condition, predict individual medication response, or recommend a dose or treatment change. Clinical medication decisions require qualified professional interpretation and appropriate confirmation. Medical content and DPYD updates need continued qualified review.
Read the analytical scope, source register, product evaluation, and security guidance.
The local server binds only to loopback, exposes public app files, rejects uploads, and sends restrictive content-security and referrer headers. To use another local port: GENEMACHINE_PORT=4174 npm start.
npm install
npx playwright install chromium webkit
npm test
npm run test:browser
npm run test:experience
npm run test:design
npm run test:appointment
npm run buildThe browser checks exercise import, technical abstention, consent, cancellation, exports, medication search, worksheet editing and ordering, combined question limits, clipboard denial recovery, responsive reflow, and requests/storage in Chromium and Playwright WebKit. The appointment suite also checks a real Chromium clipboard write/read. WebKit is not a native iPhone or Safari-device test. Synthetic fixtures test software behavior; they do not establish clinical accuracy or real-user comprehension.
Keep the default page under 220 rendered words. Evidence, file guidance, and technical checks open on demand; consent, the primary finding, source links, and a short educational boundary remain visible in their relevant flows. test:design checks this presentation and its disclosures in both browser engines.
npm run build produces dist/, containing only the static app, libraries, brand assets, synthetic example, and hosting rules. Serve it over HTTPS or localhost so workers and cryptographic hashing work. Preserve its folder structure; /web/ is the app entry. Building is separate from deployment.
The checked-in Cloudflare configuration serves static assets in Donovan's Personal account, with no database or upload endpoint. hosting/_headers preserves the local server's content-security, referrer, MIME, and cache protections. The root redirects to /web/. To release from an authorized Cloudflare login:
npm ci
npx wrangler login
npm run deployDeployment uses the pinned Wrangler version and the genemachine Worker name in wrangler.jsonc. Review the account and name before deploying a fork. No custom domain is configured. The deployed site still performs DNA parsing inside the visitor's browser.
To run the synthetic checks against a deployed HTTPS origin, set GENEMACHINE_TEST_ORIGIN to that origin and run npm run test:experience, npm run test:design, or npm run test:appointment. These check the live worker, worksheet, downloads, security headers, request origins, browser storage, mobile reflow, and blocked upload methods in both browser engines.
The browser app runs without an AI agent. The separate SKILL.md describes how an authorized agent should route genomic research through compatible tools, quality checks, current evidence, and inspected reports. The skill is a procedure, not bundled sequencing software or proof that every listed workflow is available.
git clone https://github.com/baney75/GeneMachine.git ~/.codex/skills/genemachineDifferent input classes require different tools and validation. The browser importer accepts one consumer genotype table; VCF/gVCF, BAM/CRAM, FASTQ, methylation, and multi-sample inputs require a separate reference-aware workflow. Never send a genome to an external destination without its owner’s explicit consent for that destination and purpose.
Use synthetic fixtures only. Do not submit a real genome, report, identifying filename, private record, or credential in an issue or pull request. Medical claims need dated primary sources. Tests must preserve unknown values rather than inventing reference calls. See SECURITY.md.
Licensed under Apache-2.0. Copyright 2026 Donovan Baney and contributors.