Quick Start

Go to gradiance.io and sign in with Google. The Free tier covers three runs a month with dose-response and standard-curve fitting — no card, no setup.

Click New run in the sidebar (or hit ⌘ K and pick “New run” from the command palette). You land on a setup page with three things to do, in order: upload a file, pick the assay, and (optionally) type a one-line prompt.

Drag a CSV or Excel file into the upload zone. Multi-file uploads are supported — drop several at once and they get treated as related datasets within the same run. The minimum is two columns: concentration and response, headers in row 1, numeric values below.

Concentration_uM,Viability_Percent
0.001,98.5
0.01,97.2
0.1,89.3
1,56.7
10,15.2
100,3.8

If you have replicates, repeat the row at the same concentration. Including the unit in the column name (_uM, _nM, _Percent) helps the agent label axes correctly without asking. See Uploading data for plate-reader exports and common parse errors.

The picker opens with the most likely assays at the top — Gradiance inspected the file shape and made a guess. Accept the suggestion, browse the seven categories (Quantification, Potency, Gene Expression, Kinetics, Cell-based, Binding, Custom), or use the search box if you know what you want (“ic50,” “qpcr,” “cfu”).

For 96/384-well exports the plate painteropens automatically so you can label which wells are standards, unknowns, blanks, and controls. For tabular data you can skip straight to submission.

One sentence is enough — and you can leave it blank. The agent already has the assay and the data shape; the prompt is for anything special:

Use GAPDH as the reference gene.

Variations all work: “Fix bottom to 0,” “Use the Results sheet,” “Concentrations are already log10-transformed.” On the Custom assay, the prompt is what tells the agent what kind of analysis you actually want.

Submit the form and the workspace opens — two side-by-side panels with a resizable divider, plus the agent chat as a slide-in drawer. The left panel shows the live code / progress (Pro+ sees the actual Python; Free sees a step summary); the right panel is the result viewer that fills in as the fit converges. Most runs finish in 20–40 seconds.

The viewer shows, in order:

• The fitted curve over your data points
• A parameter table with 95% CIs
• R² and a quality tier (excellent · good · fair · poor)
• Any QC flags worth your attention
• A one-paragraph plain-English interpretation

Read Understanding your results for the full tour. The short version: trust the curve before the number, and check the residuals if R² is below 0.95.

If something looks off, click Refine and tell the agent in one sentence — it refits in place without spending a new run credit.

• “Try a 5PL fit instead.”
• “Exclude the 100 µM point.”
• “Fix bottom to zero.”
• “Show me the residual plot.”

The Stop button cancels an in-progress run (failed runs don't count against your monthly quota); Duplicate clones the workflow / prompt / plot config into a new draft so you can branch on a model choice; the share icon copies the run URL to your clipboard. See The Run Workspace for the full tour of the controls.

Use the Export menu in the workspace header. Each format is self-contained.

• PNG (300 DPI) — figure for slides or print. Free tier exports include a small watermark.
• SVG — vector for editing in Illustrator or Inkscape. Pro+.
• PDF report — figure, parameters, method, provenance. Pro+.
• CSV / Excel — fitted values, residuals, outlier flags. Pro+.
• Python script — a standalone <runname>_reproducibility.py that recreates the result from raw data with NumPy + SciPy. Pro+.
• Audit bundle (JSON) — full provenance: input file hashes, fit parameters, agent steps. Pro+.