[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-83019":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":15,"subscribersCount":15,"size":15,"stars1d":16,"stars7d":17,"stars30d":18,"stars90d":15,"forks30d":15,"starsTrendScore":19,"compositeScore":20,"rankGlobal":10,"rankLanguage":10,"license":21,"archived":22,"fork":22,"defaultBranch":23,"hasWiki":22,"hasPages":22,"topics":24,"createdAt":10,"pushedAt":10,"updatedAt":28,"readmeContent":29,"aiSummary":30,"trendingCount":15,"starSnapshotCount":15,"syncStatus":31,"lastSyncTime":32,"discoverSource":33},83019,"genomi","exon-research\u002Fgenomi","exon-research","An open-source agent harness that turns your AI agent into your personal DNA expert","https:\u002F\u002Fwww.genomiagent.com\u002F",null,"Python",193,19,9,0,11,111,133,65,93.9,"Apache License 2.0",false,"master",[25,26,27],"agent-harness","genome-analysis","whole-genome-sequencing","2026-06-12 04:01:39","\u003Cp align=\"center\">\n  \u003Cimg src=\"assets\u002Fgenomi-logo.png\" alt=\"Genomi logo\" width=\"160\">\n  \u003Cbr>\n  \u003Cstrong>Your genome. Decoded.\u003C\u002Fstrong>\n  \u003Cbr>\n  \u003Ca href=\"https:\u002F\u002Fwww.genomiagent.com\u002F\">Website\u003C\u002Fa>\n  ·\n  \u003Ca href=\"https:\u002F\u002Fraw.githubusercontent.com\u002Fexon-research\u002Fgenomi\u002Fmaster\u002FINSTALL_FOR_AGENTS.md\">Install guide\u003C\u002Fa>\n  ·\n  \u003Ca href=\"README.zh-CN.md\">简体中文\u003C\u002Fa>\n\u003C\u002Fp>\n\n\u003Cp align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Fwww.python.org\u002Fdownloads\u002F\">\u003Cimg alt=\"Python\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fpython-3.10%2B-3776AB?style=flat-square&logo=python&logoColor=white&labelColor=111827\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fexon-research\u002Fgenomi\u002Freleases\u002Ftag\u002Fv0.1.0\">\u003Cimg alt=\"Version\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fversion-0.1.0-2563EB?style=flat-square&labelColor=111827\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fmodelcontextprotocol.io\u002F\">\u003Cimg alt=\"MCP\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FMCP-agent--native-7C3AED?style=flat-square&labelColor=111827\">\u003C\u002Fa>\n  \u003Ca href=\"SKILL.md\">\u003Cimg alt=\"Skill\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fskill-agent--ready-0E7490?style=flat-square&labelColor=111827\">\u003C\u002Fa>\n  \u003Ca href=\"#privacy\">\u003Cimg alt=\"Local-first\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fprivacy-local--first-15803D?style=flat-square&labelColor=111827\">\u003C\u002Fa>\n  \u003Ca href=\"LICENSE\">\u003Cimg alt=\"License\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Flicense-Apache--2.0-64748B?style=flat-square&labelColor=111827\">\u003C\u002Fa>\n\u003C\u002Fp>\n\n# Genomi\n\n> Am I going bald? What does my DNA say about Alzheimer's risk? Why does\n> ibuprofen do nothing for me?\n\nDNA is the layer underneath all of that. It shapes the proteins, enzymes,\nreceptors, and pathways behind nutrition, medication response, sleep,\nexercise, inherited traits, and risk for some conditions. Not destiny. But\nthe most personal data you carry.\n\nAnd it is overwhelming. ~3 billion base pairs, 20,000+ genes, millions of\nobserved variants per person. No clinician, no lab, no individual holds that\nin their head. It is too much.\n\nWe live in an era where AI can take on tasks that were not possible before,\nat scales never seen before. Your genome is exactly that kind of task. And\nfor the first time, we have the tools to actually read it at the scale it\nlives at.\n\nGenomi is an open-source AI agent runtime that turns your AI agent into a personal DNA expert.\nWorks with Claude Code, Codex, OpenClaw, Hermes, and any MCP-capable host. It gives the agent a private\nworkspace: your variants in a local Active Genome Index, public genetics evidence ready to\nquery, memory of what you explored, and report tools that turn DNA questions\ninto evidence-backed answers. Your genome stays on your machine. The agent\ndoes the work.\n\n## Launch video\n\n\u003Cp align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Fyoutu.be\u002F8CkoDNlyvZ0\">\n    \u003Cimg src=\"https:\u002F\u002Fimg.youtube.com\u002Fvi\u002F8CkoDNlyvZ0\u002Fmaxresdefault.jpg\" alt=\"Genomi launch video\" width=\"640\">\n  \u003C\u002Fa>\n\u003C\u002Fp>\n\n## See it in action\n\n- [Genomi parses your raw DNA file into a local database your agent can query](https:\u002F\u002Fyoutu.be\u002FmJUw6Lf8zEk)\n- [Genomi keeps your raw DNA file on your machine](https:\u002F\u002Fyoutu.be\u002FPaj2ixdeZGk)\n- [Genomi knows when to say \"No\" and \"I don't know\"](https:\u002F\u002Fyoutu.be\u002F-yXZhFDiYP0)\n- [Genomi evolves — your agent self-updates to sync with the latest research](https:\u002F\u002Fyoutu.be\u002Fih_7elp2H2w)\n\n## TL;DR\n\nEven TL;DR is too long, just paste this to your agent:\n\n```text\nHey please read this and tell me why Genomi is different from other AI\nagent harnesses. Why is this actually useful for understanding my DNA privately?\nhttps:\u002F\u002Fraw.githubusercontent.com\u002Fexon-research\u002Fgenomi\u002Fmaster\u002Fllms-full.txt\n```\n\n## Just Install It\n\nInstall it through your agent. Paste one instruction, answer a few questions,\nand let your agent wire up the runtime:\n\n```text\nInstall and configure Genomi by following the instructions here:\nhttps:\u002F\u002Fraw.githubusercontent.com\u002Fexon-research\u002Fgenomi\u002Fmaster\u002FINSTALL_FOR_AGENTS.md\n```\n\nThe install guide covers dependency checks, library selection, MCP\nregistration, optional genome-source import, and verification. If Genomi is\nalready packaged or otherwise present, the canonical install\u002Fupdate path is\n`genomi install` or the MCP operation `genomi.install`; the source bootstrap is\nonly for hosts that do not have Genomi yet.\n\n## Works With Every Agent\n\nGenomi is not tied to one chat app. Any agent host that can use MCP tools,\nlocal commands, or installed skills can talk to the same local Genomi runtime.\n\n| Host family | How Genomi connects |\n| --- | --- |\n| Claude Code | MCP server plus Genomi skills |\n| Codex CLI | MCP server plus Genomi skill |\n| OpenCode, OpenClaw, Hermes | MCP server plus host skill where supported |\n| Cursor, Gemini CLI, Cline, Goose, Roo Code, Windsurf, Claude Desktop | MCP server |\n| Any other MCP-capable host | `genomi serve` over stdio |\n\nOne local Genomi home can hold the public libraries, Active Genome Index\nrecords, score caches, and journals. Session access still follows Genomi's\napproval rules, but the underlying evidence workspace is reusable across host\nagents.\n\n## Or If You Prefer The Old-School Way\n\nClone, install, point your MCP-capable agent at it. Same flow the installer\nscript runs, just done by hand. The\n[install guide for agents](INSTALL_FOR_AGENTS.md) is the canonical reference —\nif anything below drifts from it, that doc wins.\n\n1. **Get the source.**\n\n   ```bash\n   git clone git@github.com:exon-research\u002Fgenomi.git ~\u002F.genomi\u002Fgenomi\n   cd ~\u002F.genomi\u002Fgenomi\n   ```\n\n2. **Install the package + public libraries.** The recommended install grabs\n   every default reference library so Genomi can answer real questions without\n   stopping later to fetch missing data. Use a smaller purpose from the catalog\n   only when disk, bandwidth, or time is constrained (`common-questions`,\n   `medication-response`, `ancestry-context`, `sequence-and-regions`,\n   `cell-and-tissue`, `everything`, or `setup-only`):\n\n   ```bash\n   export GENOMI_HOME=~\u002F.genomi\n   python3 scripts\u002Finstall_for_agents.py --libraries everything\n   ```\n\n   The installer creates a stable command at `$GENOMI_HOME\u002Fbin\u002Fgenomi`.\n   Add it to PATH if you want `genomi` available from any shell:\n\n   ```bash\n   export PATH=\"$GENOMI_HOME\u002Fbin:$PATH\"\n   ```\n\n   Once the `genomi` command exists, use it for install\u002Fupdate:\n\n   ```bash\n   genomi install --libraries everything\n   ```\n\n3. **Register the MCP server with your host agent.**\n\n   ```json\n   {\n     \"mcpServers\": {\n       \"genomi\": {\n         \"command\": \"\u002Fabsolute\u002Fpath\u002Fto\u002FGENOMI_HOME\u002Fbin\u002Fgenomi\",\n         \"args\": [\"serve\"]\n       }\n     }\n   }\n   ```\n\n   For a source checkout where the stable shim is unavailable:\n\n   ```json\n   {\n     \"mcpServers\": {\n       \"genomi\": {\n         \"command\": \"bash\",\n         \"args\": [\"-lc\", \"cd \u002Fpath\u002Fto\u002Fgenomi && PYTHONPATH=src python3 -m genomi serve\"]\n       }\n     }\n   }\n   ```\n\n   Reload your host's MCP servers. For URL-based ingestion, `llms.txt` is the\n   compact public map and `llms-full.txt` is one inlined reference file.\n\n## Ask It Things Like\n\nOnce Genomi is wired up, you talk to the agent like this. In Codex, use\n`$genomi` instead of `\u002Fgenomi`. The quick stuff first:\n\n> `\u002Fgenomi` What does my DNA say about Alzheimer's risk?\n>\n> `\u002Fgenomi` Am I at risk for early heart disease?\n>\n> `\u002Fgenomi` Am I going bald?\n>\n> `\u002Fgenomi` Am I a fast or slow metabolizer?\n>\n> `\u002Fgenomi` Should I worry about diabetes?\n>\n> `\u002Fgenomi` Am I lactose intolerant?\n>\n> `\u002Fgenomi` Is alcohol bad for me specifically?\n\nThen hand it something bigger:\n\n> `\u002Fgenomi` I'm about to start an SSRI. Walk me through my CYP2D6 and\n> CYP2C19 status, what the major guideline sources say about dosing, and\n> what's preliminary vs actually actionable.\n\n> `\u002Fgenomi` Run a pharmacogenomic review across every medication I take.\n> Lead with guideline-backed dose adjustments. Flag lower-evidence signals\n> second. Tell me what's outside scope.\n\n> `\u002Fgenomi` Build me a one-page rare-disease workup for my HPO terms.\n> Rank candidate genes by source-backed evidence, cite each call, and\n> show me what's missing before this is worth taking to a clinician.\n\nOr just hand it the whole thing:\n\n> `\u002Fgenomi decode`\n\nOne command. The agent sweeps every capability across your genome —\nvariants, ClinVar, pharmacogenomics, ancestry, polygenic scores,\nnutrigenomics, and your investigation journal — and serves the result\nas a self-contained dashboard on localhost. Open the URL in a browser.\n\nBehind those, Genomi gives the agent grounded tools across 20,000+ human\ngenes, millions of genotype observations from your file, and the public\nevidence sources that keep the answer honest.\n\n## What Genomi Provides\n\n| Layer | What you get |\n| --- | --- |\n| Active Genome Index | A local, queryable ledger of alleles, zygosity, quality, depth, filters, and callability context from your genome source. |\n| Evidence Library | Focused tools for variants, ClinVar, GWAS, HPO, pharmacogenomics, ancestry context, PRS, and sequence utilities. |\n| Journal | A running log of what you explored, what mattered, and which evidence supported it. |\n| Skills | Agent instructions for routing questions, asking for approval, preserving source priors, and answering clearly. |\n\n### Bringing your own genome\n\nGenomi reads your DNA from wherever it already lives. Point it at any VCF or\ngVCF you have on disk — clinical exports, research callsets, anything that\nfollows the spec — and the rest of the pipeline reuses the same Active\nGenome Index regardless of where the file came from.\n\nDirect-to-consumer providers are supported natively too. Hand Genomi the\ndeliverable straight from your account export and it figures out the rest:\n\n- **23andMe**, **AncestryDNA**, **MyHeritage**, **FamilyTreeDNA** (Family\n  Finder), and **Living DNA** — raw genotype text, zip, or `.csv.gz`.\n- **Nebula Genomics**, **Dante Labs**, and **Sequencing.com** — their VCF\n  deliverables are recognized and tagged with the originating provider.\n- **Nebula \u002F Dante \u002F Sequencing.com FASTQ** — paired-end raw reads are\n  aligned locally (minimap2 for long reads, bwa-mem2 for short reads),\n  sorted, and then fed into the same BAM → derived-VCF path. The\n  `wgs-alignment` install purpose pulls down both aligners.\n\n### No DNA file yet? Try a public one\n\nIf you don't have your own genome yet but want to see what Genomi actually\ndoes, the [Personal Genome Project — Harvard Medical School](https:\u002F\u002Fmy.pgp-hms.org\u002Fpublic_genetic_data)\npublishes real consumer-DNA deliverables from real participants. Their\ncatalog covers every provider in the list above — pick any participant's\nexport, point Genomi at it, and ask questions. It is the cleanest way to\nkick the tires without sequencing yourself.\n\nGenome data is optional; Genomi also handles public-only genetics questions.\n\n## Why We Built This\n\nI built Genomi because I want AI to take on the things it never could before,\nat the scale it never could before — and DNA is exactly that.\n\nA single human genome is overwhelming. Labs spend careers on one gene. Reports\nflatten thousands of variants into a single line. Even the best clinician\ncannot hold 20,000+ genes and millions of genotype observations in their head.\nThat is not a limitation of effort. It is a limitation of scale. And it is\nthe kind of limitation AI is finally good enough to push against.\n\nI want this for my own health. I want it for my family's health. And I want\nit to be honest — grounded in real evidence, local by default, with the agent\nshowing its work instead of guessing from memory.\n\nRaw genome files stay on your machine. Genomi is a workspace, not a static\nPDF report. Answers trace back to a source record or they don't get to call\nthemselves answers. And the whole thing is built for agents over MCP from\nthe start, not bolted on after.\n\nGeneric AI can explain genetics. It should not guess when the question\ndepends on an exact variant, your genome file, a guideline source, or a\ncoverage limitation. Genomi gives the agent the tools for the parts that\nneed evidence, and stays out of the way for the rest.\n\n## What Genomi Can Help Explore\n\nGenomi is not a static report. It is a private workspace your agent can use to\nask better questions across different parts of your genome.\n\n- Traits and everyday responses: lactose, caffeine, alcohol, taste, nutrition,\n  sleep, exercise, and similar personal questions.\n- Medication response: genes and variants that may affect how your body handles\n  specific drugs.\n- Carrier and inherited-risk context: exact variant checks, ClinVar assertions,\n  and gene-disease evidence.\n- Common-trait research: GWAS and published score context for complex traits,\n  with clear limits.\n- Rare-disease and phenotype review: HPO terms, gene-disease validity, and\n  source-backed candidate comparisons.\n- Ancestry reference-panel context: qualitative reference-panel similarity and\n  overlap checks, not race or ethnicity prediction.\n- Reports and memory: cited Markdown reports and a journal of what you explored,\n  what mattered, and what still needs follow-up.\n\n## How Genomi Keeps Answers Honest\n\nDNA questions can be personal, messy, and easy to overstate. Genomi keeps the\npieces separated so an agent can show its work.\n\n- Your genome evidence: genotype, zygosity, depth, quality, filters, exact\n  allele observation, and callability.\n- Public evidence: ClinVar assertions, population frequencies, GWAS records,\n  gene-disease validity, phenotype annotations, and source versions.\n- Reviewed findings: narrow source-backed notes recorded for a specific target\n  or question.\n- Agent memory: observations, decisions, unresolved questions, and links back to\n  evidence.\n- Personal context: optional phenotype, medications, family history, or other\n  details you choose to provide.\n\nDifferent evidence families can point in different directions. Genomi helps the\nagent compare them without pretending that one database is the whole truth.\n\n## Privacy\n\nGenomi keeps the most sensitive data close to you.\n\n- Raw genome sources stay on the user's machine.\n- Genomi creates Active Genome Index records for personal genome files locally so agents query only the\n  variants needed for the current question.\n- Genomi asks for current-session approval before read operations use existing\n  Active Genome Index artifacts, unless they belong to the configured default\n  user.\n- Public lookups use selected targets such as rsIDs, genes, drugs, conditions,\n  or guideline questions.\n- Journal entries are agent-authored memory, not evidence.\n- Project journals reject private\u002Fsample evidence links.\n- Memory exports omit private evidence links unless explicitly requested and\n  approved.\n\n## Sources, Libraries, And Attribution\n\nGenomi talks to trusted, verified databases and specialist genomics tools so\nyour agent can ground answers in real evidence instead of vibes. Install-time\ndownloads write source manifests where possible. Live adapters return source\nURLs and access context in their result envelopes. Reviewed source families are\nnot treated as background knowledge; agents cite or journal the specific source\nrecords they used.\n\nInstalled Genomi libraries:\n\n- [ClinVar](https:\u002F\u002Fwww.ncbi.nlm.nih.gov\u002Fclinvar\u002Fdocs\u002Fdownloads\u002F) —\n  `clinvar-grch38` and `clinvar-grch37` VCF caches for exact variant\n  interpretation lookup.\n- [Human Phenotype Ontology](https:\u002F\u002Fobophenotype.github.io\u002Fhuman-phenotype-ontology\u002Fannotations\u002F) —\n  `hpo` phenotype-to-gene and disease annotation files.\n- [GenCC](https:\u002F\u002Fsearch.thegencc.org\u002Fdownload) — `gencc` gene-disease\n  validity submissions.\n- [UCSC Genome Browser downloads](https:\u002F\u002Fhgdownload.soe.ucsc.edu\u002Fdownloads.html) —\n  `reference-grch38` and `reference-grch37` hg38\u002Fhg19 FASTA files for\n  sequence, normalization, and callability workflows.\n- [UCSC liftOver chain files](https:\u002F\u002Fhgdownload.soe.ucsc.edu\u002Fdownloads.html#liftover) —\n  `liftover-chains` for GRCh37\u002FGRCh38 coordinate translation.\n- [GENCODE](https:\u002F\u002Fwww.gencodegenes.org\u002Fhuman\u002F) — `gencode-grch38` and\n  `gencode-grch37` transcript annotation GTFs.\n- [ENCODE SCREEN](https:\u002F\u002Fwww.encodeproject.org\u002Fsoftware\u002Fscreen\u002F) —\n  `encode-ccre-grch38` candidate cis-regulatory element annotations.\n- [PanglaoDB](https:\u002F\u002Fpanglaodb.se\u002Fmarkers.html?cell_type=%27all_cells%27)\n  and [CellMarker 2.0](http:\u002F\u002Fbio-bigdata.hrbmu.edu.cn\u002FCellMarker\u002F) —\n  `panglaodb-markers` and `cellmarker-human` marker tables.\n- [MSigDB Hallmark](https:\u002F\u002Fwww.gsea-msigdb.org\u002Fgsea\u002Fmsigdb\u002Fhuman\u002Fcollections.jsp#H) —\n  `msigdb-hallmark`, installed only from a user-supplied official GMT export\n  or URL.\n- [PharmCAT](https:\u002F\u002Fpharmcat.org\u002F) and\n  [PharmGKB](https:\u002F\u002Fwww.pharmgkb.org\u002F) — `pharmcat` all-in-one JAR for\n  pharmacogene diplotypes, phenotypes, and recommendation artifacts.\n- [1000 Genomes 30x GRCh38](https:\u002F\u002Fwww.internationalgenome.org\u002Fdata-portal\u002Fdata-collections\u002F30x-grch38.html) —\n  `ancestry-1000g-30x-grch38` compact ancestry PCA panel, distributed from\n  the [genomi-ancestry-panel](https:\u002F\u002Fgithub.com\u002Fexon-research\u002Fgenomi-ancestry-panel)\n  build project. `ancestry-1000g-30x-grch37` is derived locally from that\n  panel with UCSC liftOver chains.\n- [minimap2](https:\u002F\u002Fgithub.com\u002Flh3\u002Fminimap2) and\n  [bwa-mem2](https:\u002F\u002Fgithub.com\u002Fbwa-mem2\u002Fbwa-mem2) —\n  `minimap2-binary` and `bwa-mem2-binary` for optional FASTQ alignment.\n  BAM\u002FFASTQ workflows also use [samtools and bcftools](https:\u002F\u002Fwww.htslib.org\u002F)\n  when those tools are needed on the host.\n\nLive public adapters and configured public data:\n\n- [gnomAD](https:\u002F\u002Fgnomad.broadinstitute.org\u002F) population frequency lookups.\n- [GWAS Catalog](https:\u002F\u002Fwww.ebi.ac.uk\u002Fgwas\u002F) association-record retrieval.\n- [PGS Catalog](https:\u002F\u002Fwww.pgscatalog.org\u002F) score metadata and scoring files.\n- [ClinPGx](https:\u002F\u002Fwww.clinpgx.org\u002F), [PharmGKB](https:\u002F\u002Fwww.pharmgkb.org\u002F),\n  [PGxDB](https:\u002F\u002Fpgx-db.org\u002F), [CPIC](https:\u002F\u002Fcpicpgx.org\u002Fguidelines\u002F),\n  and FDA [pharmacogenomic biomarker](https:\u002F\u002Fwww.fda.gov\u002Fdrugs\u002Fscience-and-research-drugs\u002Ftable-pharmacogenomic-biomarkers-drug-labeling\u002F)\n  and [pharmacogenetic association](https:\u002F\u002Fwww.fda.gov\u002Fmedical-devices\u002Fprecision-medicine\u002Ftable-pharmacogenetic-associations)\n  tables for pharmacogenomic guideline, label, and association context.\n- [KEGG](https:\u002F\u002Fwww.kegg.jp\u002Fkegg\u002Fpathway.html),\n  [Reactome](https:\u002F\u002Freactome.org\u002F),\n  [QuickGO](https:\u002F\u002Fwww.ebi.ac.uk\u002FQuickGO\u002F),\n  [Human Protein Atlas](https:\u002F\u002Fwww.proteinatlas.org\u002F), and\n  [ChEMBL](https:\u002F\u002Fwww.ebi.ac.uk\u002Fchembl\u002F) for pathway, ontology,\n  tissue\u002Fcell-type, compound, and drug-target relationships.\n- [Open Targets Platform](https:\u002F\u002Fplatform.opentargets.org\u002F) for disease and\n  clinical drug-target context.\n- [BioGRID ORCS](https:\u002F\u002Forcs.thebiogrid.org\u002F),\n  [DepMap](https:\u002F\u002Fdepmap.org\u002Fportal\u002Fdownload\u002F), and\n  [NCBI GEO](https:\u002F\u002Fwww.ncbi.nlm.nih.gov\u002Fgeo\u002F) for configured or discovered\n  functional-genomics perturbation evidence.\n\nReviewed source families:\n\n- [ClinGen Gene-Disease Validity](https:\u002F\u002Fsearch.clinicalgenome.org\u002Fkb\u002Fgene-validity),\n  [GeneReviews](https:\u002F\u002Fwww.ncbi.nlm.nih.gov\u002Fbooks\u002FNBK1116\u002F),\n  [MONDO](https:\u002F\u002Fmondo.monarchinitiative.org\u002F),\n  [Orphanet](https:\u002F\u002Fwww.orpha.net\u002F), [OMIM](https:\u002F\u002Fwww.omim.org\u002F),\n  [GeneCards](https:\u002F\u002Fwww.genecards.org\u002F), [MalaCards](https:\u002F\u002Fwww.malacards.org\u002F),\n  [NCI cancer genetics resources](https:\u002F\u002Fwww.cancer.gov\u002Fabout-cancer\u002Fcauses-prevention\u002Fgenetics),\n  and the [COSMIC Cancer Gene Census](https:\u002F\u002Fwww.cosmickb.org\u002Fknowledgebase\u002Fcosmic-modules\u002F)\n  are source families agents may review, cite, and journal for disease,\n  cancer-risk, and gene-context investigations.\n- [DrugBank](https:\u002F\u002Fgo.drugbank.com\u002F),\n  [PharmaProjects](https:\u002F\u002Fpharmaintelligence.informa.com\u002Fproducts-and-services\u002Fdata-and-analysis\u002Fpharmaprojects),\n  and [PubMed](https:\u002F\u002Fpubmed.ncbi.nlm.nih.gov\u002F) support reviewed\n  drug-target, mechanism, and primary-literature context when the agent records\n  specific source-backed findings.\n\n## How It Works\n\nGenomi exposes a small base MCP surface plus a dispatcher for specialized\ngenomics tools. The host agent does the conversation; Genomi does the grounded\nlookup, Active Genome Index creation, evidence retrieval, and report assembly.\n\n1. Connect an agent over MCP — see [the install steps](#or-if-you-prefer-the-old-school-way)\n   above for the config snippet.\n\n2. Give the agent a genome file when you want personal context.\n\nGenomi parses the file into an Active Genome Index: a local query substrate for\nvariants, zygosity, quality, depth, filters, and callability context. Public-only\nquestions do not require a genome file.\n\n```json\n{\n  \"tool\": \"genomi.parse_source\",\n  \"params\": {\n    \"source\": \"\u003Cgenome-file>\"\n  }\n}\n```\n\n3. Ask questions. The agent calls the smallest useful Genomi operation.\n\nBase operations such as `genomi.parse_source`, `genomi.describe_context`, and\n`journal.append_entry` are direct MCP tools. Capability operations go through\n`genomi.invoke` after the agent reads the matching `skills\u002F\u003Ccapability>\u002FSKILL.md`.\n\n```json\n{\n  \"tool\": \"genomi.invoke\",\n  \"params\": {\n    \"tool\": \"variant.resolve\",\n    \"params\": {\n      \"rsid\": \"rs429358\"\n    }\n  }\n}\n```\n\n4. Inspect evidence, defaults, and limitations.\n\nGenomi results include structured evidence, source coverage, and\n`defaults_applied` where assumptions matter. Missing libraries, unavailable\nexternal sources, and background jobs return explicit statuses instead of being\ntreated as negative evidence.\n\n5. Remember.\n\nThe Journal records observations, decisions, unresolved questions, and evidence\nlinks.\n\n## Build With Genomi\n\nGenomi is open source and built for people who want AI agents to work with\ngenomics responsibly: local-first, evidence-grounded, and honest about\nlimitations. Use it to explore, explain, remember, and report on DNA questions\nwithout uploading the raw genome file.\n\n## Status\n\n> [!WARNING]\n> **Experimental. Research and informational use only.**\n> Genomi is not a diagnostic device. It does not replace qualified clinical\n> review for diagnosis or treatment. Raw genome data stays on your machine\n> by design — but you are still responsible for how you share what comes out\n> of it.\n\nThe schema, tool surface, and capability layout are still moving — pin a\ncommit if you need stability across upgrades.\n\n## License\n\nGenomi is released under the [Apache License 2.0](LICENSE).\n\n## Citation\n\nIf you use Genomi in research, publications, reports, benchmarks, demos, or\nderived tools, please cite the project using [CITATION.cff](CITATION.cff) and\nacknowledge Genomi where appropriate.\n\n```bibtex\n@software{genomi2026,\n  title = {Genomi: A Local Genomics Harness for AI Agents},\n  author = {Zeng, Mingde and Zhou, Hongjian and Liu, Fenglin and Wu, Jinge},\n  year = {2026},\n  url = {https:\u002F\u002Fwww.genomiagent.com\u002F},\n  version = {0.1.0}\n}\n```\n\n## Contributing\n\nIssues and pull requests welcome. If you are reporting a bug, include the\ngenome source format (VCF \u002F gVCF \u002F 23andMe \u002F AncestryDNA \u002F etc.), the\noperation you ran, and the structured error envelope the agent received —\nthat is usually enough to reproduce.\n\n## Acknowledgements\n\nGenomi owes a direct implementation debt to the\n[Personal Genome Project — Harvard Medical School](https:\u002F\u002Fmy.pgp-hms.org\u002Fpublic_genetic_data)\npublic genetic data catalog.\n\nThat same PGP-HMS public dataset also did the unglamorous work of letting\nGenomi support all of these providers natively. Every detector, every\ncolumn quirk, every header banner, and every test fixture for the\nconsumer-array and provider-tagged VCF paths was sanity-checked against\nreal PGP participant exports. Native MyHeritage, FamilyTreeDNA, Living\nDNA, Nebula, Dante, and Sequencing.com support exists because PGP-HMS\nmakes real-world examples freely available under a permissive re-use\nlicense — a quiet contribution to open consumer genomics that Genomi\ninherits directly.\n\nThanks also to [GBrain](https:\u002F\u002Fgithub.com\u002Fgarrytan\u002Fgbrain), Garry Tan's\nOpenClaw\u002FHermes agent-brain project, for inspiration around making agent\nsystems source-grounded, memory-aware, and useful from a single fetched\ndocumentation entry point.\n","Genomi 是一个开源的AI代理运行时，能够将你的AI代理转变为个人DNA专家。它支持Claude Code、Codex、OpenClaw、Hermes等MCP兼容主机，并为代理提供了一个私有工作空间，包括本地活动基因组索引中的变体、可查询的公共遗传学证据、探索记忆以及将DNA问题转化为基于证据的答案的报告工具。其核心功能在于解析原始DNA文件并创建本地数据库供代理查询，同时确保所有数据均存储在用户设备上，以保障隐私安全。Genomi适用于希望深入了解自身基因信息及其对健康影响的个人，也适合研究人员或医疗保健专业人士用于辅助分析复杂的基因数据。",2,"2026-06-11 04:09:53","CREATED_QUERY"]