Frequently asked.

Yes — Oreumics is free during the beta and no credit card is required. Access is granted through a free, admin-issued Beta Access ticket that unlocks the full feature set; request one via the contact form.

Beta Access is admin-issued and time-limited (typically 7, 14 or 30 days) and unlocks the whole product — all 119 nodes and 33 templates across 5 modalities (scRNA-seq, Spatial, Bulk RNA-seq, Methylation, ATAC-seq), the AI Copilot and agent, assistant connections over MCP, reports, Figure Canvas and multi-workflow tabs. There is no reduced tier.

Go to /contact and choose “Beta access request” as the inquiry type. Tell us briefly who you are, your institution, and what you would like to analyse. We review every request manually and email you a ticket code that you redeem from the dashboard. We prioritise academic labs, postdocs and researchers with a concrete project in mind.

Your local files, .bioflow projects and saved workflows are not affected — they are on your disk, not on our servers. To keep using the app, request a new Beta Access ticket; it is free during the beta.

Yes. BioFlow Copilot became Oreumics in v0.25.0 (September 2026) — same app, same team, same local-first design. Projects, .bioflow files, sign-in and tickets carry over unchanged; the installer, window titles and Start Menu entry now say Oreumics. Compatibility identifiers such as the .bioflow extension were deliberately kept.

After the beta. Billing will run through Paddle, and every active user will be notified by email before any paid plan takes effect. There is no Lite tier and there will not be one.

Not yet. Oreumics currently runs on Windows 10/11 (64-bit) only. macOS support is on our roadmap; tell us at gptapikorea@gmail.com if you need it, because demand from beta users decides the priority.

Linux is not supported in the current beta. It may be added later based on user feedback.

Windows 10/11 64-bit, 8 GB RAM minimum (16 GB recommended), disk space for the app with its bundled Python engine plus a working cache, and an internet connection at every launch for sign-in. We do not publish a cell count per amount of RAM: how far a dataset goes depends on how it is stored and which steps you run, and the app predicts that per node before it starts.

Yes. Oreumics is not yet code-signed, so Windows Defender SmartScreen may warn. Click “More info” → “Run anyway”. Code signing is planned to remove the warning.

The installer is per-machine — it installs Oreumics for all users of the PC — so Windows shows its administrator approval (UAC) prompt. Click Yes. This is separate from the SmartScreen warning above; on a managed PC, ask your IT administrator to run the installer.

Use the contact form with “Report a problem” as the inquiry type (/contact?type=bug) and paste the relevant part of backend.log. The app writes its logs to %APPDATA%\BioFlow Copilot\logs — backend.log for the analysis engine, main.log for the app shell — and the Log panel's “Open log folder” button takes you there (the folder name is a compatibility identifier from before the rename). Skim the log first and remove any file paths you consider sensitive: it records the paths of the files you loaded. During the beta we reply within 2 business days.

After sign-in the app checks for a newer version. If there is one, an update window offers to download and install it now, with a progress bar, or to start the current version and update later. Updates never touch your projects.

Yes. Every analysis step (Scanpy, Squidpy, PyDESeq2, mepylome, decoupler and the rest) runs locally, and your sequencing data and AnnData files are never uploaded. What does use the network: the sign-in check at launch, update checks, one-time Resource Center installs (which send only a selector such as a version or organism), the AI Copilot if you configure a key (node names, parameters and small result summaries — never raw matrices), and the online Enrichr service if you use it for pathway enrichment (it receives your gene list; a local gene-set library avoids it).

It runs the full analysis locally — it is not a viewer. QC, normalisation, PCA, clustering, DEG, trajectory and RNA velocity, CellRank, spatial statistics, bulk DESeq2, methylation DMP / DMR, ATAC and cell–cell communication all execute on your machine and write real results you can export as H5AD, CSV, PNG / PDF / SVG, or a report.

22 loader nodes. Single-cell: Load H5AD, Load 10X H5, Load 10X MTX, Load CSV / TSV, Load Loom, Load Text, Load Velocyto Loom, Load STARsolo Velocity. Spatial: Visium (SpaceRanger folder), Visium HD (binned), Xenium, MERSCOPE, CosMx, transcript tables and CSV coordinates. Bulk: count matrices, featureCounts / STAR ReadsPerGene, Kallisto abundance.tsv and Salmon quant.sf. Methylation: Illumina IDAT (27K / 450K / EPIC v1 / EPIC v2 / Mouse) and beta matrices. ATAC: fragments, peak matrices and Cell Ranger ATAC MEX.

No sample data ships in the installer, and templates open with the loader path empty. The Sample datasets page lists the public data each template was built around and where to download it.

Since v0.25.0 a new workflow tab first asks what you are analysing — single-cell RNA-seq, spatial, bulk, methylation or ATAC — and the Node Library then shows only that modality’s nodes, in pipeline order from loader to plot. Search always covers every node, and “All nodes” restores the classic category layout.

Yes. Every plot node opens into a Visualization Workspace with the same five sections (Plot · Titles & Labels · Axes · Legend · Figure & Export): per-group colours that follow the group into every later plot, group renaming, a legend placement grid, font sizes, transparent background, PNG / PDF / SVG export, and Export raw data for the numbers behind the figure. Re-generate redraws only the figure.

Yes — merge samples with Merge Data (which keeps a per-sample label), then run Harmony on the PCA space and point Neighbors at X_pca_harmony. Annotation Concordance helps judge whether labels agree across methods afterwards.

Some nodes need something that is not your data — a CellPhoneDB database, a CellTypist model, the OmniPath ligand-receptor network, PROGENy / Hallmark / CollecTRI priors, JASPAR motifs, methylation manifests, ENCODE cCRE regions, a genome FASTA. The Resource Center installs them once, pinned by version and checksum with licence acceptance, and a run never downloads anything. A template that contains such a node stops at that node with a message naming the resource until it is installed; the other nodes still run.

The dialog says what is holding the memory and what to change. The usual answers, in order: re-run the loader if the message mentions a dense matrix (v0.25.0 converts dense-on-disk data to sparse on load); keep “Scale HVG Only” on; run Highly Variable Genes before PCA / Neighbors; close other applications; switch the memory profile to Low memory; or Subset Data first. Run Anyway overrides the prediction at the risk of the watchdog aborting the task. The Memory & large data guide walks through each.

Raw-signal preprocessing lives in Load IDAT’s prep parameter: raw, illumina, noob or swan (SWAN is not deterministic). Methyl Normalize offers NaN-aware quantile normalisation or none. BMIQ and noob were removed from Methyl Normalize in v0.14.4 because the implementations behind those labels were not what they claimed and corrupted beta values silently; results made with them should be re-run.

Yes. Export CSV writes 15 tabular result types; every visualization workspace exports the table behind its figure with a README; Save H5AD writes the full AnnData; figures export as PNG, PDF or SVG; Figure Canvas exports composed multi-panel figures; and reports come out as HTML or PDF. A .bioflow project export carries run receipts and an RO-Crate provenance record.

Same inputs, same parameters and the same seeds give the same result for deterministic nodes, and every run records its parameters, seeds and library versions. Two things are not deterministic by nature and are labelled as such: SWAN preprocessing, and anything that asks a language model.

OpenAI (gpt-5.6-terra by default), Anthropic (claude-opus-5 by default) and Google Gemini. You bring your own API key, set in the chat panel’s API Key Settings; Oreumics does not mark up model costs. Without a key the app works fully — only the Copilot, LLM-based annotation and report narrative need one.

No. The Copilot is optional and, when used, sends only lightweight metadata — node names, parameters and the small summaries or tool results the model asks for — never raw counts. Auto Annotation sends cluster marker lists. If you use the online Enrichr service for pathway enrichment, your gene list is sent there; a local gene-set library avoids it.

Only visibly. It proposes a plan card that you Apply (and can Undo in one step); in Auto mode safe plans apply automatically, but deletions and disconnections always wait for you. Approve & Run lets it apply, run, verify against the run receipts and re-plan until a goal is met, with Stop and Undo available throughout.

Claude Desktop, the ChatGPT desktop app (Codex) and Claude Code connect over MCP through the in-app “Connect an AI assistant” dialog: Claude Desktop via Save extension file… and drag-and-drop, the other two via config files the app writes for you. ChatGPT on the web or phone cannot connect, because it calls out from the cloud and cannot reach your PC. Plans from a connected assistant arrive on the canvas as the same Apply cards.

A sign-in window opens every time you launch Oreumics, and the app verifies your access with the server before it starts. If your session from a previous launch is still valid the window passes straight through without asking you to type anything — but it always appears, and it always checks. Sign out from the window to switch accounts.

No — the app cannot start without a connection, because sign-in is verified at every launch; offline, Oreumics reports a connection error and closes. Once running, the analysis itself is fully local. If you need an air-gapped machine, tell us and we will discuss an offline licence.

Not with the same account at the same time. One account is bound to one active device: signing in on a second machine ends the first session, and that app shows a warning and closes after a short grace period so you can save. Sign in again on whichever machine you want to keep using.

Contact us at gptapikorea@gmail.com and we will delete your account and associated data within 7 business days. Your local projects are unaffected because they were never uploaded.