THE SYSTEM

An evidence-first AI system for life sciences.

Claude composes a shared reasoning layer with specialized DomainOS products over an MCP data layer. Claims are grounded in artifacts. Artifacts carry provenance. Critical decisions route through human gates.

4 reasoning plugins · 24 DomainOS · 6 planned · 10 MCP data servers

THE CO-WORK SYSTEM

How a request flows.

A request passes through 6 stages — intent routing, plugin dispatch, parallel agents, evidence verification, human gates, canonical artifacts — with typed contracts at each boundary.

REQUEST FLOW

01

User intent

Intent classification with a confidence score
02

Engine dispatch

Routing contract, HITL level, evidence mode
03

Parallel execution

DomainOS + MCP + KG (validated | exploratory | degraded)
04

HITL gates

Hypothesis, design, execution, publish checkpoints
05

Evidence Guardian

Grounding + provenance + contradiction detection
06

5 canonical artifacts

result, report, provenance, gate_status, session

SYSTEM LAYERS

ORCHESTRATOR

hordago core

Intent router + evidence contracts

MCP SERVERS

10 servers

PubMed, ChEMBL, bioRxiv, Knowledge Graph…

REASONING PLUGINS

4 composed plugins

co-writer, co-scientist, paper-intel, knowledge-graph

DOMAINOS

24 specialized domains

CRISPRos, GWASos, BIOos, ONCOos…

THE REASONING LAYER

4 plugins Claude composes.

Each plugin is an evidence-first reasoning system with its own skills, agents, and MCP tools. Claude composes them together for any workflow — writing, hypothesizing, reasoning, tracking claims, connecting knowledge.

co-writer

Compiles scientific writing

Compiles submission-grade papers, grants, biosketches, and CVs from structured intent and lab data. Every output compiled from artifacts, with claim-to-evidence tracking and venue formatting.

— compile-manuscript
— evidence-audit
— scientific-writing
— peer-review
— compliance-lint

co-scientist

Generates and critiques hypotheses

Hypothesis-driven research assistant. Generates, critiques, and evaluates hypotheses using the Swanson ABC model across a 6-phase pipeline — Scout → Generate → Critique → Scan → Plan → Review — with Elo tournament ranking.

— hypothesis-generate
— hypothesis-critique
— analysis-plan
— novelty-scan
— gap-detect

paper-intelligence

5 MCP

Extracts and tracks claims

Extracts structured claims from scientific papers, builds evidence graphs, and queries claim relationships. Seven-stage ingestion: ingest → extract → normalize → dedup → link → tag → upsert. Every claim carries the source it came from.

— paper-ingest
— claim-extract
— claim-graph-query
— contradiction-detection
— claim-provenance

knowledge-graph

11 MCP

Unifies biomedical knowledge

Universal biomedical knowledge graph infrastructure. Neo4j graph + DuckDB columnar + SQLite FTS5 text search, Biolink Model schema, namespace-isolated across pi, t1d, gwas, crispr, tools, hyp, skill.

— graph-query
— graph-ingest
— graph-search
— graph-explore
— graph-status

DOMAINOS PRODUCTS

One stack, every life-science domain.

Each DomainOS packages the workflows, evidence gates, and provenance requirements of its field, and composes the reasoning plugins for that domain's motions.

Genomics & Variants (9)

Cell & Molecular (5)

CRISPRos

Guide design, knockouts
SCREENos

CRISPR screens, MAGeCK
CELLos

Cell biology analysis
SPATIALos

Spatial transcriptomics
ATLASos

Cell atlas construction

Omics (6)

Imaging & Structure (2)

STRUCTos

AlphaFold, docking
BIOIMAGEos

Microscopy

Clinical & Regulatory (2)

RAREos

Rare disease, ACMG
CLINICALos

EHR and cohort analysis

Drug Discovery (3)

CHEMos

Cheminformatics
ADMETos

ADMET prediction
VIRTUALSCREENos

Virtual screening

Systems & Networks (3)

SYSTEMSos

Systems biology
PATHWAYos

Pathway analysis
GRNos

Gene regulatory networks

MCP DATA LAYER

Ten servers, on demand.

On-demand access to biomedical databases via Model Context Protocol. Servers sit idle until a tool is invoked — only the query and its response consume context.

biocontext7

Deep bioinformatics skill bundles

stdio

PubMed

36M+ biomedical articles

HTTP

ChEMBL

2.4M bioactive compounds

HTTP

bioRxiv

Preprints (bio + med)

HTTP

ClinicalTrials

ClinicalTrials.gov registry

HTTP

Open Targets

Drug target evidence

HTTP

Synapse

Sage Bionetworks data

HTTP

BioRender

Scientific figure generation

HTTP

Knowledge Graph

Neo4j biomedical KG

stdio

Paper Intelligence

Claim graph queries

stdio

COMPOUND WORKFLOW CATALOG

6 capability tiers of typed skill contracts.

Every skill is a typed contract with inputs, outputs, provenance, and validation. Agents compose them without inventing steps.

Writing & Publication

34

manuscripts, grants, CVs, presentations, posters, slides, lit reviews, peer review, Google Docs

Discovery & Hypothesis

8

co-scientist pipeline, hypothesis-generate/critique, novelty-scan, gap-detect, controversy-map

Knowledge & Claims

17

paper ingest, claim extraction, graph queries, gap detection, contradiction scan, citation networks

Statistics & Analysis

9

Cox regression, power-analyst, multi-comparison, effect-size, assumption-checker, test-selector, time series

Data Visualization

11

matplotlib-publication, seaborn-stats, interactive-viz, figure-compose, schematics, citation networks

Verification Infrastructure

6

verified-skills (CONTRACT.json), evidence-audit, context-discovery-mcp (GPU embeddings)

BY THE NUMBERS

Current state.

24

DomainOS with a repository

6

DomainOS planned

7

Scientific Domains

10

MCP Data Servers

4

Engine Plugins

30

Registry Entries

30

Registry Entries

An evidence-first AI system for life sciences.

How a request flows.

User intent

Engine dispatch

Parallel execution

HITL gates

Evidence Guardian

5 canonical artifacts

4 plugins Claude composes.

co-writer

co-scientist

paper-intelligence

knowledge-graph

One stack, every life-science domain.

Ten servers, on demand.

6 capability tiers of typed skill contracts.

Current state.

Compose the whole system.