Oreumics turns 119 analysis steps into a visual graph on a Windows desktop. Drag a loader onto the canvas, branch into trajectories, spatial statistics or motifs, and walk the same canvas to a PDF report — with an AI Copilot that can edit the graph and an agent that runs it, step by step, in front of you.
scRNA-seq · Spatial · Bulk RNA-seq · Methylation · ATAC-seq. Every node computes locally; the app signs you in at launch.
Every analysis path — single-cell, spatial, bulk RNA-seq, ATAC-seq, methylation — walks the same five stages on the same canvas, with the same property panel and the same workspaces. The nodes differ; the way you work does not.
Load
22 loaders. H5AD, 10x H5 and MTX, CSV/TSV, Loom, Velocyto and STARsolo for single-cell; Visium, Visium HD, Xenium, MERSCOPE, CosMx and transcript tables for spatial; featureCounts, Kallisto and Salmon for bulk; IDAT and beta matrices for methylation; fragments and peak matrices for ATAC. Merge Cell Meta attaches a sample sheet.
QC
The Quality Control node opens a QC Workspace: MAD-based threshold recommendations, a Raw Data Mode for unfiltered droplets, and thresholds you move live without re-running. Scrublet flags doublets. Bulk QC, Methyl QC, ATAC QC and Spatial Density QC cover the other modalities.
Analyse
PCA, Harmony, neighbors, UMAP and t-SNE; Leiden or Louvain community detection on the kNN graph, or K-Means with a fixed number of clusters on the embedding plus K-Means; DEG, Enrichr and GSEA; 3 activity nodes; 4 annotation nodes; CellPhoneDB and LIANA; PAGA, DPT, scvelo and CellRank; squidpy spatial statistics, BANKSY domains and cellular neighborhoods; PyDESeq2; DMP, DMR and GOmeth; LSI and JASPAR motifs. Change one parameter and only the downstream nodes re-run.
Visualise
18 plot nodes draw 93 plot types. Each opens a Visualization Workspace with the same five sections, and every finished figure lands in the Figure Canvas gallery for layout, annotation and PNG/SVG/PDF export. The raw data behind any figure exports as CSV with an .LLM_README.md.
Report
The report outline is detected from your graph, you set the emphasis in a short chat, and Oreumics writes HTML or PDF. A Methods paragraph is generated from the nodes and parameters you actually ran, and project export carries run receipts and an RO-Crate.
Each card links to the page that describes the feature as it ships in v0.25.0.
Interactive views that open from a node card and change the figure or the data without re-running the pipeline.
Set thresholds on the real distributions, then apply them once.
One vocabulary of controls shared by every plot node.
Every finished figure in one gallery, laid out by hand.
Browse and tidy cell metadata before it drives an analysis.
From graph to HTML or PDF in three steps.
119 nodes in 14 categories, organised by modality.
Optional AI that reads node names, parameters and small summaries — never a matrix — and a receipt for every run so nothing has to be taken on trust.
Chat that reads your data and edits your graph.
Approve once; the agent runs and verifies each step on your canvas.
Every run leaves a receipt.
The parts of the app that hold a project together: the canvas, the templates, the resources and the memory guard.
Several workflows and Figure Canvases side by side, in split panes.
33 templates across 5 modalities, with sample data.
External databases and models are installed before a run, never during one.
Peak memory is predicted per node before it runs.
Save, open, export and import as a single .bioflow file.
One installer for Windows 10/11; sign in at every launch.
Electron front end, a bundled Python engine behind it on localhost. Your counts, IDATs and Xenium folders are read from disk and stay on disk. Windows 10/11, 64-bit.
Scanpy, Squidpy, PyDESeq2, scvelo, CellRank, decoupler, LIANA, CellPhoneDB, CellTypist, harmonypy, Scrublet, gseapy, Hotspot, mepylome and pyJASPAR, called through nodes instead of a notebook. BANKSY spatial domains and cellular neighborhoods are native implementations.
Nodes form a DAG. Change a parameter and only the downstream nodes are invalidated and re-run; cached results reopen from a saved project; a running node can be cancelled. Large results live outside the node card so the canvas stays light.
Every node runs on your machine. The app verifies your account at launch, so it needs a connection to start, and the optional AI Copilot sends node names, parameters and small summaries only. External resources download once, from the Resource Center, before a run.
Counts are read from the node catalog that the app generates — 892 parameters in total, documented per node under Docs → Nodes.
BioFlow Copilot is now Oreumics
Same app, same local-first design, a new name and mark. Projects, .bioflow files, sign-in and access tickets carry over unchanged.
Connect an AI assistant, in-app
A guided dialog replaces the separate window. Claude Desktop gets a Save extension file… step (the Microsoft Store edition cannot open .mcpb by double-click); the Codex app and Claude Code have their config written for them; the dialog confirms the first real request.
Runs that fit in memory are admitted
Nodes are charged only for the matrices they read, the disk cache quota is a cleanup target instead of a wall, dense matrices are converted to sparse on load, and Normalize and Log1p share buffers.
Resource Center follow-ups
A missing database stops the node with a message naming the resource, CellPhoneDB databases installed through the Resource Center are recognised, and the AI can ask which resources are installed before proposing a plan.
Spatial viewer and canvas fixes
Polygon selection, per-group highlighting and a zoom that survives re-rendering; large tables no longer cut off mid-stream. Multi-select Ctrl+C works again and Merge Cell Meta resolves auto-match the same way in preview and run.
Reproducible Draw Graph and a modality-first library
Force-directed layouts are seeded and report the layout actually used. The Node Library is organised by modality first.
Free during the beta — access by admin-issued ticket. No credit card required.