Summary

Genomi is an open-source AI agent runtime that transforms AI agents into personal DNA experts. It tackles the overwhelming scale of human genomic data by providing a local, private workspace for agents to query and interpret an individual's DNA. This enables users to explore genetic predispositions, medication responses, and inherited traits with evidence-backed answers, without uploading sensitive raw genome files.

How It Works

Genomi parses a user's raw DNA file (e.g., VCF, 23andMe export) into a local "Active Genome Index," a queryable database of genetic variants. AI agents, connected via MCP, use this index and a curated library of public genetics evidence (ClinVar, GWAS) to ground their analysis. This approach enables detailed exploration of traits, pharmacogenomics, and rare diseases while maintaining user privacy and allowing agents to self-update with the latest research.

Quick Start & Requirements

Installation is typically handled via an agent's command interface or a provided script (scripts/install_for_agents.py). Manual installation involves cloning the repository and running the script. Genomi requires an MCP-capable host agent (e.g., Claude Code, Codex CLI). Optional dependencies like minimap2/bwa-mem2 may be needed for FASTQ alignment. The genomi install --libraries everything command installs default libraries. Further setup involves registering the Genomi MCP server with the host agent.

Highlighted Details

• Privacy-Centric: Raw genome data remains local.
• Agent-Native: Built for AI agents via MCP.
• Evidence-Grounded: Responses traceable to specific genomic sources.
• Broad Data Support: Handles VCF, gVCF, and major DTC genetic testing provider exports.
• Local Active Genome Index: Efficient variant querying.

Maintenance & Community

The project welcomes issues and pull requests, with key contributors including Mingde Zeng and others. Specific community channels are not detailed, but active development is indicated by evolving schema and APIs.

Licensing & Compatibility

Released under the Apache License 2.0, which permits commercial use, modification, and distribution, making it compatible with most integration scenarios.

Limitations & Caveats

Marked "Experimental" for research/informational use only; not a diagnostic device. Users are responsible for output privacy. The schema and API are subject to change, requiring commit pinning for stable deployments.