Graphomics logoGraphomicsRead the Docs

General models invent biology. We build the layer they check against.

Graphomics is the verified data plane for biological AI — a curated knowledge graph, validated pipelines, and a provenance ledger, callable over MCP from the tools your team already uses. Runs managed, self-hosted, or licensed — your data never has to leave your perimeter.

Read the DocsTalk to Us →
MapForgeWorkbenchNexus
TAXAMETABOLITESDISEASE
The Problem

What scientists were promised — and what they actually get.

Promised

— Gene–disease links, proposed on demand
— Mechanistic hypotheses built from omics data
— Investigation cycles run end to end
— Publication-ready reports drafted in minutes

Reality

— Species, diseases, and pathways get hallucinated
— No source of truth linking gene → disease → target
— 50+ bioinformatics tools that don't connect
— No data isolation between one lab and the next

The reasoning is already a commodity. What's missing isn't intelligence — it's a trustworthy substrate to reason over.

The Insight

AI got good at code because code can be checked.

An agent writing software has somewhere to stand: it can run the code, read the trace, watch the test fail, and correct itself. Biology gives an agent none of that. There is no executable, inspectable record to ground against — so it guesses. Build that record, and the agent becomes trustworthy.

01
Executable
An answer becomes an operation that can be run and re-run, not a paragraph of text.
02
Inspectable
Every intermediate step is exposed as structured, queryable evidence you and the agent can audit.
03
Stateful
Progress persists across steps — a memory of what was decided that survives the loop.
The Loop

Ground. Execute. Contribute.

Ask a grounded question, reason over the graph, run a validated pipeline — then write the result back with its provenance. That return step is what closes the loop.

The Ground

MapForge

Builds a production knowledge graph from any biological data — ingests sources, links entities, scores confidence on every connection, serves it through an API.

Learn more →
The Hands

Workbench

50+ validated bioinformatics pipeline steps that actually run the analysis. Multi-tenant, RBAC, per-org storage, MCP-native.

Learn more →
The Mind

Nexus

A research agent that reasons over the graph and orchestrates the pipelines, in natural language.

Learn more →

↻ Every answer writes back to the graph — with its provenance.

MCP

Built on open standards. Any agentic tool — Claude Code, Cursor, your own — plugs in over MCP. Nexus is included, not required.

Deployment

Runs where your data already lives.

Managed, self-hosted inside your own environment, or licensed outright. Your proprietary data never leaves your perimeter.

Managed
We run it. You query it. Nothing to maintain.
Self-hosted
Deployed inside your own environment, under your controls.
Licensed
License the engine outright and run it anywhere.
Proof

MicroMap isn't the product. It's proof the engine works.

MapForge built MicroMap — a microbiome knowledge graph — from raw sources in days, no manual curation pipeline.

63,316
microbe–disease associations, spanning 2,981 microbes and 243 diseases
1.7M+
entities
1.1M+
taxa
1,400+
diseases
6,500+
metabolites
6,200+
drugs
2,028
proteins
60+
API endpoints
50+
analysis nodes
Who It's For

Three ways in.

Platform team

Point MapForge at your own data and license the engine that builds the graph.

Explore MapForge →
Lab or core facility

Run validated pipelines with real isolation between projects and teams.

Explore Workbench →
Bench scientist

Ask a grounded question in plain English, get a cited answer. No coding required.

Explore Nexus →
Why Now

Three things changed.

01 · Capability

Reasoning got good, and cheap. Frontier models can plan an experiment and read an analysis. Intelligence is no longer the bottleneck.

02 · Standards

MCP turned every pipeline and data source into something an agent can call. The plumbing is finally open.

03 · Stakes

A hallucinated pathway still costs months at the bench. Grounding isn't a nicety here — it's the whole job.

The reasoning arrived this year. The ground to reason over didn't.

Every lab. Every organism.
Every answer, grounded.

Read the DocsTalk to Us
Follow alongSubstack ↗X ↗