Skip to main content

FOR LIFE SCIENCES RESEARCH · LABS · BIOTECH · GOVERNMENT · INDUSTRY

Traceable claims. Reproducible figures. Auditable experiments.

Evidence-first AI infrastructure for life sciences. Claude composes 24 Domain Operating Systems, a shared reasoning layer and an MCP data layer — with human gates at every critical decision.

Book a Demo How it works
Powered by Claude Awardee Anthropic AI for Science
hordago-cli
Synthetic data Placeholder inputs and figures, shown to explain the workflow.

Built for life sciences research

Academic labs
Biotech
Pharma & industry
Government
Clinical genomics
Core facilities
Research hospitals
Contract research

The system

One orchestrator. Every life-science domain.

Claude composes specialized reasoning plugins with a biomedical knowledge graph, a claim extraction layer, and an MCP data plane — with human gates at every critical decision.

24

Domain Operating Systems

each with a repository · 6 more planned

7

Scientific domains

genomics · omics · clinical · drug discovery

4

Reasoning plugins

composed by Claude on every run

10

MCP data servers

zero context cost until queried

Counted from the product registry in this repository

The approach

The compiler for biology.

Claude orchestrates the run — reading intent, proposing specs, composing specialized agents, drafting prose. The Evidence Guardian verifies every output against source artifacts, with human gates on every critical decision. The result is a reproducible manifest you can audit end-to-end.

A bench protocol that always runs the same way — your data in, validated figures out.

Input

Data + specs

CSV · FASTQ · VCF · YAML

Orchestrate

Claude + agents

intent · spec · review

Verify

Evidence Guardian

grounding · provenance · HITL

Output

Reproducible artifacts

figures · stats · manifest

Built for trust

Human governance by default.

Agentic AI is powerful. Science demands accountability. Every Hordago product enforces human oversight at decision points.

Human approval gates

AI proposes. Scientists decide. Critical decisions route through explicit sign-off.

Full provenance

Inputs, commits, tool versions, environment, SHA-256 hashes — everything recorded in a reproducible manifest.

Grounded claims

Outputs are compiled from artifacts. Every sentence references a concrete figure, statistic, or citation.

See it in action

What a run produces.

Every CRISPRos run produces traceable artifacts with full provenance. The panel below is a hand-written illustration of a guide design workflow, not a capture of one.

CRISPRos — Guide Design Output 
$ hordago design-guides --gene BRCA1 --modality knockout --species human

┌─────────────────────────────────────────────────────────────┐
│  CRISPRos v0.8.2  ·  Guide Design Pipeline                 │
└─────────────────────────────────────────────────────────────┘

✓ Genome loaded: GRCh38 (hg38)
✓ Gene: BRCA1 — chr17:43,044,295-43,125,364
✓ Exons scanned: 23 coding exons

── Guide Candidates ──────────────────────────────────────────

  Guide 1:  ATCGATCGATCGATCGATCGTGG   Efficiency: 0.89
  Guide 2:  GCTAGCTAGCTAGCTAGCTAAGG   Efficiency: 0.84
  Guide 3:  TTAACCGGTTAACCGGTTAACGG   Efficiency: 0.81
  Guide 4:  CCGGAATTCCGGAATTCCGGGGG   Efficiency: 0.77

── Off-Target Analysis ───────────────────────────────────────

✓ 0 off-target hits in coding regions
✓ 2 intronic near-misses (3+ mismatches) — flagged, not blocking

── Provenance ────────────────────────────────────────────────

✓ manifest.json written  (sha256:a3f8c2...)
✓ Git commit: e4b7d1f
✓ All inputs + outputs hashed and recorded

→ Output: guides/BRCA1_knockout.json
→ Status: VALIDATED — ready for human approval
Synthetic data Placeholder inputs and figures, shown to explain the workflow.

Claude orchestrates

Delegate the workflow. Keep command.

Claude reads a plain-language request, dispatches the right reasoning plugins, composes specialized agents, and routes data through the MCP layer — with human gates at every critical decision. You keep the final call on hypothesis, design, execution, and publish.

Synthetic data Placeholder inputs and figures, shown to explain the workflow.

watch the real Claude Cowork substrate ↗

The request loop

Four stages from plain-language intent to a reproducible artifact — each gated by evidence contracts and human approval.

Built on the Claude Code and Agent SDK foundation. Runs against Claude with extended thinking, structured output, and parallel tool use.

  1. 01

    Intent classification

    Claude reads a plain-language request and routes it across the 13 Hordago task categories with a confidence score.

  2. 02

    Plugin dispatch

    A routing contract selects the right mix of co-writer, co-scientist, paper-intelligence, and knowledge-graph plugins for the job.

  3. 03

    Parallel agent execution

    Specialized agents run in parallel — each with its own skill set, evidence contract, and HITL checkpoint.

  4. 04

    Evidence Guardian review

    A grounding + provenance + contradiction sweep before results leave the orchestrator.

See the full 4-tier architecture Book a demo to see it live

Domain Operating Systems

30 products. One stack.

Purpose-built operating systems for every life-science domain. Each packages its field's workflows, evidence gates, and provenance requirements, and composes the reasoning plugins that fit those motions.

Genomics & Variants

9 products

  • BIOos

    Beta

    Alignment, QC, RNA-seq

  • GWASos

    Beta

    Association, fine-mapping

  • VARFXos

    Beta

    Variant annotation

  • FINEMAPos

    Beta

    Statistical fine-mapping

  • PRSos

    Beta

    Polygenic risk scores

  • MENDELos

    Beta

    Mendelian genetics

  • CLINVARos

    Beta

    ClinVar interpretation

  • LONGREADos

    Beta

    ONT, PacBio

  • REGos

    Beta

    Enhancers, TF binding, chromatin

Cell & Molecular

5 products

Omics

6 products

  • ONCOos

    Beta

    Somatic, neoantigen

  • MICROBIOMEos

    Beta

    16S, metagenomics

  • PROTEOMICSos

    Beta

    Mass spec

  • METABos

    Beta

    Metabolomics

  • IMMUNOos

    Beta

    Immune repertoire

  • MULTIos

    Beta

    Multi-omics integration

Imaging & Structure

2 products

  • STRUCTos

    Beta

    AlphaFold, docking

  • BIOIMAGEos

    Beta

    Microscopy

Clinical & Regulatory

2 products

  • RAREos

    Beta

    Rare disease, ACMG

  • CLINICALos

    Planned

    EHR and cohort analysis

Drug Discovery

3 products

  • CHEMos

    Beta

    Cheminformatics

  • ADMETos

    Planned

    ADMET prediction

  • VIRTUALSCREENos

    Planned

    Virtual screening

Systems & Networks

3 products

  • SYSTEMSos

    Planned

    Systems biology

  • PATHWAYos

    Planned

    Pathway analysis

  • GRNos

    Planned

    Gene regulatory networks

Explore the architecture Request a domain

24 Domain Operating Systems

BIOos GWASos VARFXos FINEMAPos PRSos MENDELos CLINVARos LONGREADos REGos CRISPRos SCREENos CELLos SPATIALos ATLASos ONCOos MICROBIOMEos PROTEOMICSos METABos IMMUNOos MULTIos STRUCTos BIOIMAGEos RAREos CHEMos BIOos GWASos VARFXos FINEMAPos PRSos MENDELos CLINVARos LONGREADos REGos CRISPRos SCREENos CELLos SPATIALos ATLASos ONCOos MICROBIOMEos PROTEOMICSos METABos IMMUNOos MULTIos STRUCTos BIOIMAGEos RAREos CHEMos

STAY IN THE LOOP

Get early access updates

Join researchers and developers who get the latest on evidence-first AI for life sciences. Product launches, science deep dives, and early access invitations.

No spam. Unsubscribe anytime. We respect your inbox.

Bring evidence-first AI to your lab.

Hordago Labs works with research labs, biotechs, government programs, and life-science industry teams. Claude orchestrates the specialized agents behind every workflow, with human gates on every critical decision and a full provenance manifest on every output.

Book a Demo Read the architecture