Local. Provable. Complete.
The moat is three words. Your analysis never leaves your machine. Every run is a visible node graph with a receipt you can audit and reproduce. And 5 modalities — scRNA-seq, Spatial, Bulk RNA-seq, Methylation, ATAC-seq, with a no-code methylation path from raw IDAT — live in one desktop tool. Not a cloud you rent. Not a black box you trust. Not a single assay you outgrow.
- 119
- nodes
- 14
- categories
- 5
- modalities
- 100%
- local
Each pillar is a strength you can rely on.
Your data never leaves the machine.
Every analysis — Scanpy, Squidpy, PyDESeq2, methylation arrays — runs on your own hardware through a bundled Python engine. Raw counts, patient samples and AnnData files stay on disk. The app checks your account when it starts; that sign-in is the only thing that needs a connection. The AI Copilot is optional and sends only lightweight metadata (node names, parameters, small result summaries), never your raw data, and can be left switched off.
Kills — cloud lock-in
Cloud-only tools force your sequencing data onto someone else’s server before you can touch it — a non-starter for clinical, IRB-restricted, or unpublished data.
A graph you can see, audit, and re-run.
The pipeline is a visible node graph — not a scrolling chat log or a wall of generated code. Click any node to see its parameters, change one, and only the downstream nodes re-run. Every run leaves a receipt with its parameters, seed and determinism class, and the project export carries those receipts as an RO-Crate. Same inputs, same parameters, same seed give the same result for deterministic nodes; the few steps that are not deterministic — SWAN preprocessing, LLM annotation — say so in the receipt instead of pretending.
Kills — black-box AI & hallucination
When an agent answers in prose or emits code you can’t verify, you can’t tell a correct result from a confident-sounding wrong one. A graph makes the method inspectable.
5 modalities, one no-code desktop.
scRNA-seq, Spatial, Bulk RNA-seq, Methylation, ATAC-seq — 119 nodes across 14 categories and 22 loaders, in a single installer with its own Python engine. No conda, no R, no stitching five tools and five file formats together. Methylation goes from raw IDAT to DMP, DMR and GOmeth on the same canvas as your single-cell work.
Kills — single-modality & vendor lock-in
Point tools cover one assay and leave the rest to R scripts or a second vendor. Breadth in one tool means your methylation, spatial, and single-cell work share one canvas.
Knows your machine’s limits before a run
The memory guard predicts each node’s peak before it runs, admits or refuses it with a plain-language message naming what is holding memory, re-measures near-misses instead of refusing on a single snapshot, and watches the run live. You find out before a step would exceed your workstation — not halfway through it.
You’re always in control
Cancel stops a running node or pipeline mid-flight. Every AI edit is Apply / Undo — a plan you approve, never a silent mutation of your canvas — and the agent’s Approve & Run has a Stop button and a step budget.
Small labs get three bad doors. Oreumics is the fourth.
Door 1
Upload to someone’s cloud
Your unpublished or patient data leaves the building before you can analyze it.
Door 2
Burn a grad student for six months
Turn a scientist into a scripting technician, then hope the pipeline is reproducible.
Door 3
Trust a black-box AI
Take generated code or prose on faith, with no way to see or check the method.
The fourth door: a no-code tool you install once, that runs on your own machine, and that shows its work as a graph you can edit and reproduce.
Where Oreumics wins — and why.
9 concrete strengths, no spin — the things you can do on day one, from raw files to publication-ready results, entirely on your own machine.
No-code methylation from raw IDAT
5 modalities in one tool
100% local analysis
Full analysis pipeline (not just a viewer)
No-code / 0-setup
AI that proposes, you approve
Vendor-neutral
Annotation you can cross-check
Transparent pricing
We’d rather leave a slot empty than fake it.
These are reserved for real, named design-partner stories with quantified before/after outcomes — not stock quotes or invented logos. For a privacy-first, radically honest tool, an honest placeholder beats fake social proof.
““6 weeks of R scripting → 2 days, fully reproducible.””
Small academic wet-lab
Named user + institution + one quantified before/after
“Analysis stayed inside the building — nothing uploaded.”
Clinical / privacy-first group
Named user + institution + privacy outcome
“IDAT → DMR → GOmeth, no bioinformatician on staff.”
Design-partner case study
One paper or project that used Oreumics end to end
The gaps we’ll tell you about.
If another tool is a better fit for your specific need, we’ll say so. Credibility comes from honesty — here is where Oreumics is partial, roadmap, or deliberately out of scope.
Windows only
Oreumics ships for Windows 10/11, 64-bit. There is no macOS or Linux build.
Sign-in at every launch
The app verifies your account each time it starts, on one active device, so it will not open without a connection. The analysis itself needs none.
No proprietary cell atlas
We don’t ship a curated reference atlas of our own. If your workflow depends on a vendor atlas, that’s a gap today.
No reference-atlas / Census auto-annotation
Excluded on purpose: the required native dependency has no cross-OS wheel, so it wouldn’t run everywhere. We support CellTypist, CellMapper (kNN), and LLM annotation instead.
No VDJ / immune-repertoire
TCR/BCR repertoire analysis is not in scope yet.
No CITE-seq / protein workflow
There is no node for antibody-derived tag counts; only the gene-expression matrix of a multimodal dataset is analysed.
Local AI clients only
Claude Desktop, Claude Code, the Codex app and Codex CLI connect over MCP today. Cloud-hosted clients — the ChatGPT app, Claude on the web or on a phone — cannot reach a desktop app.
The agent runs on your canvas, not headless
Approve & Run drives the visible workflow in front of you, which is what makes every step reviewable and undoable. It is not a server-side runner: closing the app stops the run, and there is no unattended CI mode.
Local · Provable · Complete
The bioinformatics core your lab can own.
Raw single-cell, spatial, methylation, ATAC & bulk data → publication-ready, reproducible results with a receipt for every run. No code. No cloud. Free beta — access by request.