Vol. 01 · No. 1 · Beta Edition
Oreumicsbeta

●Single-cell · Spatial · ATAC · Bulk · Methylation

Describe the analysis. Watch it build. Edit any step.

Oreumics is the AI copilot for single-cell, spatial & multi-omics bioinformatics. Say what you want in plain language — it assembles a visual pipeline of 119 analysis nodes1, runs it on your machine2, and lets you open any node to tune it. Publication-ready figures and reports, without a line of code.

Start freeDownload appFree beta · access by request · Windows 10/11
119
nodes
5
modalities
22
loaders
100%
local
The actual workspace — a live pipeline
Oreumics — pbmc_analysis.bioflow
▶ Run
Load H5AD

pbmc_3k.h5ad

QC Filter

min_genes: 200 · MT < 5%

Normalize

target_sum: 1e4

UMAP

n_neighbors: 15

Clustering

resolution: 0.5

Nodes appear, wire together, then run — ending in a UMAP. The desktop app and this site share one design system.

  1. 01.The AI Copilot does not just answer questions. It returns an Action Plan — a structured list of node insertions, parameter updates, and edge changes — that you stage with one click. See §02.
  2. 02.Analysis runs 100% locally on your machine. The optional AI Copilot sends only lightweight metadata — node names, parameters, and small result summaries, never your raw data — to your chosen LLM provider, and can be turned off entirely.
How it works

From a sentence to a result — in three moves.

The AI proposes, the canvas makes it visible, and you stay in control. Ask in plain language, watch the pipeline build node by node, then open any step to tune it — nothing hidden, nothing you can’t change.

01Ask

Describe it in plain language

The AI answers with a staged Action Plan — add nodes, wire them, set parameters — that you approve with one click.

02See

Watch the pipeline build

Every step lands as a visible node on the canvas, wired in order. Nothing runs in a black box.

03Edit

Open any node and tune it

Change a parameter and only downstream re-runs — deterministic, logged, and reproducible.

Output

“The supplement writes itself.”

Methods, figures, statistics, and AI-written interpretation — exported as HTML or PDF, ready for the supplementary materials.

Oreumics

scRNA-seq Analysis Report

PBMC 3k Dataset · February 25, 2026

OreumicsAI-Assisted Analysis

🔬 Executive Summary

Analysis of 2,638 high-quality cells following stringent QC identified 8 transcriptionally distinct populations via Leiden clustering (resolution=0.5). T cell clusters (C3, C7) are characterized by high CD3D and CD3E expression, with C7 corresponding to CD8+ cytotoxic T cells (GZMB↑, PRF1↑). A monocyte cluster (C4) demonstrates elevated LYZ and CST3. B cells form a compact cluster (C5) defined by MS4A1.

Contents

•Data Overview
•Quality Control
•Clustering Analysis
•Differential Expression
•Methods
•AI Insights

🔵 Clustering Analysis

UMAP 1UMAP 2

Fig. 1: UMAP projection

Leiden clustering produced 8 populations ranging from 89 to 480 cells. Cluster C1 (n=480) represents CD4+ helper T cells, while C3 (n=312) are NK cells. The separation in UMAP space reflects distinct transcriptional programs confirmed by differential expression analysis (p < 0.001).

Generated by Oreumics · Feb 25, 2026

Export PDF
Share
See it in action

The actual screens — not stock art.

Chat builds and edits the graph, your own AI assistant can drive it over MCP, and every figure is tuned in a live workspace. Here is what that looks like.

01 · AI Chat

Chat that knows your data.

Ask in plain language and the AI answers from your actual analysis — it reads the clusters and marker genes and tells you which are T cells, which are monocytes. It can also add nodes and edit parameters, staged as a diff you approve with one click.

Oreumics — AI Chat
Oreumics AI Chat answering which clusters are T cells versus monocytes, grounded in the real marker genes from the analysis — Cluster 0 CD4 T cells (CD3D/CD3E), Cluster 2 classical monocytes (S100A8/S100A9/LYZ)
Oreumics — AI Chat
Oreumics AI Chat staging an Action Plan: create a doubletDetection node and wire it between Load and Quality Control, with green Apply and outlined Approve & Run buttons — nothing runs until you approve

02 · Action Plan

Ask for a change — approve the diff.

The AI stages every edit as an Action Plan — add a node, wire it in, tune parameters. Nothing touches your workflow until you press Apply.

PreviewOr Approve & Run — the agent applies it, runs it, and verifies each step, with one-click undo.

Claude Desktop
Claude DesktopCodexClaude Code● Oreumics connected
Run QC and clustering on my open PBMC workflow in Oreumics.
I’ll read your open workflow, then propose a plan for the canvas:
oreumicslist_workflows() · get_workflow_graph()
pbmc_3k.bioflow · 4 nodes · revision 12
oreumicspropose_canvas_plan(run: qualityControl → clustering, mode: auto)
Applied on the canvas · 2,638 / 2,700 cells pass QC · 8 clusters
Done — 8 clusters found. The results are on your canvas.
Oreumics MCP server · localhost — your data never leaves the machine

03 · MCP

Preview

Or drive it from your own AI.

Connect Claude Desktop, Codex, or Claude Code over MCP. They call Oreumics’ tools on your machine — running nodes, reading results — and your data never leaves the device.

04 · Figure workspace

Tune every figure, live.

Recolour, resize, relabel, swap colormaps — in an interactive workspace. Every tweak is saved back to the node, so the figure stays reproducible.

Oreumics Embedding Workspace — tuning a PBMC UMAP figure live: colormap, point size, opacity, and per-cluster colours
Goal-seeking agent

Tell it the outcome — it finds the settings.

Say “cluster this until I get exactly 20 populations.” The agent picks a resolution, runs clustering, counts the result, and re-plans — looping until it lands on the number. You approve once; it stops at the goal, a token budget, or your Stop button.

  • ✓Verified, not claimed — every iteration is checked on the canvas — the cluster count is read back from your data, not asserted in chat.
  • ✓It can’t run away — a finite token and step budget caps the loop; it stops at the goal or the budget, whichever comes first.
  • ✓Fully reversible — one-click Undo reverts the whole run, and Stop halts it mid-loop.
How the agent loop works
Oreumics — Agent · goal-seek↻ Looping
Cluster this until I get exactly 20 populations.
🎯 Goal — Leiden clusters = 20
#1resolution0.50
#2resolution1.10
#3resolution0.85
plan → run → verify → re-plan · Stop anytime

clusters found

—

target20

Illustrative animation — the goal-seeking loop itself is in the shipping app.

Single-cell RNA-seq

Full scanpy pipeline, visually.

From raw counts to publication-ready figures. Each step is a node you can connect, configure, and re-run instantly — no kernel restart, no boilerplate.

  • Quality control & doublet removal
  • Normalisation, HVG, scaling
  • PCA, UMAP, t-SNE
  • Leiden / Louvain clustering
  • Trajectory analysis (scVelo, PAGA)
  • Harmony batch correction

Pipeline · PBMC

  1. 01Load H5AD2,700 cells
  2. 02Scrublet62 doublets removed
  3. 03QC Filter2,638 cells pass
  4. 04Normalizelog-normalised
  5. 05HVG select2,000 genes
  6. 06PCA50 components
  7. 07UMAP2D embedding
  8. 08Clustering8 clusters

Figures generated by Oreumics

Fig. 1 — 3

Fig. 1 · UMAP

UMAP · Cell clusters

UMAP 1UMAP 201234567

2,638 cells · leiden clustering

Fig. 2 · Violin

Violin · CD3D expression

TNKBMonoNK2CD3DCD3D by cell type

T cell marker expression

Fig. 3 · Dot plot

Dot plot · Marker genes

C1C2C3C4C5CD3DCD8AMS4A1LYZPPBPsize = % expressinglowhighexpr

Top markers per cluster

Spatial transcriptomics

Decode tissue architecture.

The full Squidpy stack on a visual canvas. Understand how cells are organised in tissue — and which molecules drive it. Compatible with 10x Visium, MERFISH, Xenium, Slide-seq.

  • Spatial neighbours graph
  • Neighbourhood enrichment
  • Spatially variable genes
  • Co-occurrence & Ripley statistics
  • CellPhoneDB ligand–receptor
  • Spatial scatter & domain plots

Pipeline · Visium

  1. 01Read Visium3,460 spots
  2. 02Spatial neighborsn_rings: 1
  3. 03Nhood enrichmentp < 0.001
  4. 04Spatial SVGs243 genes
  5. 05Spatial scattertissue map

Spatial Scatter · 10x Visium

spatial 1spatial 2
Layer 1
Layer 2
Layer 3
Core
Vessel
Beyond single-cell

One canvas, every modality.

Bulk RNA-seq

9 nodes

featureCounts, STAR, Kallisto or Salmon in; PyDESeq2 differential expression, GSEA, volcano and MA plots, sample-distance heatmaps out.

DNA methylation

12 nodes

Illumina IDAT or beta matrices. Raw-signal QC, DMP/DMR, CpG-bias-corrected GOmeth, and local cCRE/BED enrichment.

ATAC-seq

9 nodes

Fragments or peak matrices. ATAC QC, peak calling, LSI, differential accessibility, motif enrichment via pyJASPAR.

Pathways & activity

5 nodes

Enrichr ORA and gseapy GSEA for gene lists; PROGENy, Hallmark and CollecTRI activity scores per cell or sample.

Cell–cell communication

4 nodes

CellPhoneDB + LIANA consensus ligand–receptor analysis. Dot plots, heatmaps, chord networks.

● Unique to Oreumics

The only no-code path for methylation arrays.

Load raw IDAT / 450K / EPIC / EPIC v2, normalize, call DMPs & DMRs, and run GOmeth pathway enrichment — on a canvas. No R, no scripting, no local toolchain to wrangle.

Explore methylation

The pipeline · 12 nodes

  1. 01Load IDAT
  2. 02Methyl QC
  3. 03Quantile normalize
  4. 04DMP Analysis
  5. 05DMP feature enrichment
  6. 06Methyl GOmeth
Capabilities digest

Everything a researcher needs — nothing more.

01

Node-based workflow

Drag, drop, connect. No code, no Jupyter, no kernel restarts.

02

AI copilot built in

Chat proposes editable action plans; open any node to tune it. Bring your own OpenAI, Claude, or Gemini key.

03

Publication-grade figures

UMAP, violin, dotplot, spatial scatter, volcano — exported as high-res PNG.

04

Runs 100% locally

Your AnnData never leaves the machine. Scanpy + Squidpy under the hood.

05

DAG execution engine

Parallel branches, downstream cache invalidation, real-time progress.

06

Auto-updating desktop app

New features ship over the air. Always on the latest build.

And an AI copilot that understands every one of them.

Read the article

Subscribe to the beta

Start your analysis— today.

Small labs get three bad doors: upload to someone’s cloud, burn a grad student for six months, or trust a black-box AI. Oreumics is the fourth — the bioinformatics core your lab can own.

Free during beta · no credit card · local execution · admin-reviewed access.

Create free accountView pricingWhy Oreumics