exon-research/genomi

Genomi – an AI‑agent‑powered personal DNA assistant

What it is – Genomi is an open‑source runtime that lets a local AI agent (Claude Code, Codex, OpenClaw, Hermes, etc.) act as a private genomics expert. It builds a local “Active Genome Index” from your raw DNA files (VCF, gVCF, consumer‑genotype CSVs, or even FASTQ reads) and wires the agent up to a curated set of public genetics databases. The agent can then answer health‑, trait‑, or ancestry‑related questions while keeping the raw genome on your machine.

Why it matters – Human genomes contain millions of variants; no person or clinician can memorize them. Genomi gives an AI the ability to look up exact variant evidence, pharmacogenomic guidelines, GWAS scores, and more, on‑demand, without sending any personal data to the cloud. It’s a local‑first “workspace” rather than a static PDF report, so the agent can cite sources, track what you’ve explored, and refuse to answer when evidence is missing.

---

Core components

Layer	What you get
Active Genome Index	A fast, queryable ledger of every allele in your genome (zygosity, depth, quality, filters, callability).
Evidence Library	Pre‑downloaded adapters for ClinVar, gnomAD, GWAS Catalog, PharmGKB, ClinGen, OMIM, KEGG, Reactome, and dozens of other curated resources.
Journal	An auto‑generated log of questions, answers, and the exact evidence records cited.
Skills	MCP‑compatible command definitions that let any supported agent invoke Genomi (/genomi …).

---

How you use it

1. Install – Run the provided scripts/install_for_agents.py (or the genomi install shortcut) which pulls the default libraries (everything) and creates a genomi command in ~/.genomi/bin.
2. Register with your agent – Add a MCP server entry pointing to genomi serve (or the shim) in the host’s configuration JSON.
3. Load your DNA – Point Genomi at a VCF/gVCF, a consumer‑genotype CSV (23andMe, AncestryDNA, etc.), or a FASTQ pair (it will align locally with minimap2/bwa‑mem2 and call variants).
4. Ask questions – In the chat UI use commands like:

   /genomi What does my DNA say about Alzheimer’s risk?
   /genomi Run a pharmacogenomic review of my meds.
   /genomi decode   # full dashboard of traits, scores, ancestry, etc.
   

   The agent replies with a concise answer, citations, and optionally a Markdown report.

---

Privacy by design

• Local‑first – Raw genome files never leave your computer.
• Session approval – Genomi asks for explicit permission before reading the Active Genome Index for a new session.
• Separate evidence – Public look‑ups are limited to the specific rsIDs/genes you query; the journal never stores private variant links unless you explicitly allow it.
• Export control – Memory exports strip private evidence unless you opt‑in.

---

Who might find it useful?

Audience	Typical use‑case
Health‑curious individuals	Quick, evidence‑backed answers about disease risk, drug metabolism, or nutrition based on their own genome.
Clinicians / genetic counselors (as a research aid)	Rapidly explore a patient’s variant list, pull ClinVar classifications, and generate a draft report that can be reviewed.
Developers of AI agents	Plug‑in a ready‑made genomics skill set via MCP without building the data pipelines themselves.
Researchers	Test pipelines on public genomes (e.g., PGP) and prototype new phenotype‑gene queries.

---

Ecosystem & Extensibility

• MCP‑compatible – Works with any host that can launch an MCP server (genomi serve).
• Skill files – SKILL.md defines the /genomi commands; agents can extend them.
• Library catalog – Install a subset of evidence (common-questions, medication-response, ancestry-context, etc.) to save space or bandwidth.
• Public‑library updates – The runtime can self‑update its evidence caches to stay current with the latest ClinVar releases, GWAS catalog, etc.

---

Quick start snippet (from the README)

# Clone and install
git clone https://github.com/exon-research/genomi.git ~/.genomi/genomi
cd ~/.genomi/genomi
export GENOMI_HOME=~/.genomi
python3 scripts/install_for_agents.py --libraries everything
export PATH="$GENOMI_HOME/bin:$PATH"

# Register with your agent (example JSON)
{
  "mcpServers": {
    "genomi": {"command": "$GENOMI_HOME/bin/genomi", "args": ["serve"]}
  }
}

After registration, simply type /genomi … in your chat interface.

---

Bottom line

Genomi turns a generic LLM‑based chat agent into a private, evidence‑grounded genomics assistant. It handles the heavy lifting of parsing raw DNA files, maintaining a local variant index, and exposing a rich set of curated genetics databases—all while keeping your personal genome on‑device and giving you transparent, cited answers.