GitHubLinkedInStack OverflowEmailBuy me a coffee

gramp — your genome, analyzed privately

Encrypted on-device, interpreted with LLMs and live biomedical literature. Raw genotypes never leave your machine.

The idea

Consumer DNA kits hand you a raw text file and a walled-garden web report. The interesting question — what does the current research actually say about my variants? — means uploading the most sensitive data you own to someone else's server. gramp answers it without that trade: it imports your raw export, encrypts it at rest, and does the interpretation locally, calling out to biomedical literature and an LLM of your choice while keeping the genotypes themselves on your laptop.

What it does

  • Import raw DNA — 23andMe, AncestryDNA, MyHeritage exports, or VCF. Every genotyped position is indexed by rsID, chromosome, and locus, and stored in an encrypted database.
  • Traits & persona — a GWAS-based trait catalog classifies markers as good / watch / neutral, and an LLM writes a plain-language summary of your genetic profile.
  • Live research — scheduled scans query Europe PMC (PubMed plus bioRxiv/medRxiv preprints) against your trait markers, surfacing new publications as findings. Research any variant on demand.
  • Health context — import an Apple Health export and gramp keeps only 30-day rolling aggregates (min/max/avg), never the raw time series, to give the analysis longitudinal context.
  • Documents & chat — drop in PDFs or notes; gramp embeds them for local vector search and folds them into a conversational analysis over your traits, health, and files.
  • Reports & export — daily/weekly/monthly reports, plus selective or full encrypted backup of variants, traits, findings, and documents. Your data stays portable.

Privacy is the architecture, not a setting

The database is encrypted at rest with a passphrase-derived key (Argon2), optionally unlocked from the OS keychain. Genotypes never leave the device — when the analysis needs the cloud, gramp assembles a de-identified context (rsIDs and aggregated health passages only) and logs exactly what left the machine in an egress audit. Point it at a local Ollama model with on-device embeddings and it runs fully offline, no internet required.

Bring your own model

gramp is vendor-agnostic: Anthropic Claude, OpenAI GPT, or local Ollama, with a quick mode for fast summaries and a deep mode for careful analysis. API keys live in the OS keychain, never in plain text. Retrieval runs over PubMed and bioRxiv today, with an MCP framework for plugging in other research sources.

Status

gramp is an early-stage desktop app (Electron, React, TypeScript), built local-first from the ground up. The core flows — import, encrypted storage, trait interpretation, scheduled research, health context, and chat — are wired end to end. It's a personal project and not a medical device: nothing it surfaces is diagnosis or medical advice. Interested in early access or the engineering behind it?Reach out.