Genomi

Agent harness that turns your agent into personal DNA expert

Health & Fitness

Analytics

Biohacking

GitHub

Hey folks! I want to introduce Genomi, an agent harness that I've been building for a while and dogfooding it along the way. I think it's an incredible time to be building in this space. We finally have powerful agent hosts running right on our machines, things like Claude Code, Codex, OpenClaw, and Hermes Agent, they have completely change how we work. Like a lot of people, I took a DNA test years ago. I remember getting the report, found something mildly interesting, and immediately forgot about it. It just sat in a zip file on my hard drive. Recently, I tried giving that data to an AI agent to ask some health and genetic context questions. It was mediocre at best. The current agent tools simply cannot handle a raw VCF or large genotype file. If you try to link it in the agent, the sheer volume of data instantly blows up the context window, or the agent must read it line by line, and it is still overwhelmingly error-prone. There are two other problems. Static DNA reports can't keep up with new science. They're out of date the moment they're generated. And your DNA data should stay on your own device. No one should have to upload deeply personal, non-rotatable genomic data to some startup's website just to analyze it, especially with all the privacy concerns and bankruptcies piling up in the consumer testing space (looking at you, 23andMe). So we built Genomi. It's a local-first, agent-native, evidence-grounded harness that uses the MCP and SKILLs to bridge the gap between raw genomic data and LLMs without choking your agent environment. Tools like Claude Code and Codex route their LLM inference to the cloud by default, so I designed Genomi specifically to handle the context size and the data exposure. Your raw DNA file never leaves your machine. Genomi parses it locally into an air-gapped, queryable database on your own hardware, called the Active Genome Index. The genome itself stays put. And yes, your agent's own LLM still sees the questions you ask and the findings it pulls back, so if you want zero data leaving at all, you can pair Genomi with an agent environment running on a local model fully offline. Because genetics research moves quite fast, running /genomi update syncs your agent's local workspace with the latest research releases, so your evidence base never goes stale. To stop the agent from leaning on hallucinations, Genomi gives it 88 tools wired into roughly 30 public genetics databases like ClinVar, gnomAD, PharmCAT, CPIC, and the FDA tables. It forces the agent to inspect real scientific evidence and show its work, and respond in confidence levels. So what does it actually feel like to use it? You can query specific things via your agent chat: > `/genomi` Am I a fast or slow metabolizer? > `/genomi` Will I go bald? > `/genomi` Why does ibuprofen do nothing for me? Or you hand it the whole genome at once with /genomi decode. It sweeps every capability across your DNA, variants, ClinVar, pharmacogenomics, ancestry, polygenic scores, the works, and serves it as a self-contained dashboard on localhost. This is still experimental and at the early stage, we are eager to hear any feedback for y'all, the project is released under Apache 2.0 so feel free to play around with it, and join us in making it better! GitHub: https://github.com/exon-research... Website: https://www.genomiagent.com/