Nonlinear regression, peak fitting, hypothesis testing and phase identification for XRD, spectroscopy, kinetics and mechanical data. Every fit reports parameter uncertainties, and every result is checked against the reference Python libraries before release.
Free to use · Runs in your browser · Your data stays on your computer
Built for
Data in one program, fits in another, figures in a third, and a script nobody else in the group can run.
One workspace in your browser: paste the data, get the fit, statistics, publication figure and methods text in minutes.
See what it analyses →A fit with too few points, an ANOVA on skewed data, a crystallite size beyond what the method can measure.
An engine checked against SciPy, and Facet, an AI lab partner that reviews every analysis for these mistakes before you publish.
How the numbers are checked →Which phase is in this pattern? Which composition, recipe or temperature should the next run use?
Phase identification against 5.8 million computed crystals, Bayesian optimisation and generative design point to the next experiment.
See discovery and AI →Lattent computes every fit, test, model and figure in your browser. Unpublished measurements, formulations and process data are not uploaded to be analysed, which is why R&D groups that cannot send data to cloud tools can use it.
The analysis engine and all sixteen models are JavaScript running on your machine. Heavy jobs run in a background thread of the same browser tab.
Open your browser’s network panel while you work. Apart from loading the page, no request carries your data unless you choose one of the options below.
Projects reach our database only when you press Save to cloud. Row-level security limits access to you and your lab, and you can delete everything at any time.
Facet’s suggestions, checks, graph reading and reports run locally. The optional chat goes directly from your browser to the provider you choose, under your own account.
Database searches in Discover send only element names or formulas. For an internal installation behind your firewall, talk to us. Details: privacy notice · security.
Language models are superb at explaining and planning. They are not a dependable way to produce the numbers in your paper.
Microsoft’s MatterGen and Google DeepMind’s GNoME showed a new way to find materials: generate candidates, screen them with machine-learned property models, and check them against computed data. Lattent brings that loop to your bench at the level of chemical composition, next to the 5.8 million DFT structures of the open Alexandria database.
Each material comes with its structure in 3D and conventional cell, formation energy, distance to the convex hull, band gap, a simulated powder pattern with Miller indices, and a CIF file.
From a peak fit to a ranked list of phases in seconds. Every candidate cell is symmetrised, simulated and matched with a refined lattice scale; peaks left unexplained point to a second phase.
Lattent’s own masked-diffusion model, trained on the Materials Project, proposes formulas in your chemical system, steered towards a band gap. Property networks predict formation energy (MAE 0.09 eV/atom) and band gap, and the database shows which candidates are new.
Honest limits: Lattent generates compositions, not full crystal structures, and its predictions are leads for a DFT calculation or a synthesis, not results. Lattent is not affiliated with Microsoft, Google DeepMind, the Materials Project or the Alexandria project.
Lattent understands the measurements materials labs run every day. Fit overlapping XRD, Raman or XPS peaks on a fitted baseline; get d-spacings, Scherrer crystallite size and Williamson–Hall strain; extract optical band gaps from UV–Vis; and read modulus, yield, strength and toughness from stress–strain curves.
Thirty-five equations from materials science, polymers, nanomaterials and chemistry (Arrhenius, Langmuir and Sips, Avrami, Carreau, WLF, Hall–Petch, Tafel and more), or type your own. Lattent's shape model recognises what kind of behaviour your data show and narrows the equations to the plausible ones. Auto-fit then ranks them by AICc, so you see what the data support.
Lattent learns a Gaussian-process model of a property from the experiments you have run, such as the toughness of a polymer blend, the yield of a synthesis or the conductivity of an electrolyte. It shows which inputs matter and proposes the next experiments most likely to improve the result, respecting mixture constraints. Add them to your table, run them, and repeat.
t tests, Mann–Whitney, one-way ANOVA, Welch's ANOVA and Kruskal–Wallis, with Tukey and Holm–Šidák post tests. Lattent checks equal variance and normality for you, suggests the right test when an assumption fails, and draws significance stars on the graph.
Open Facet, Lattent’s AI lab partner, on any table. Its built-in suggestions come from Lattent's own shape model, a neural network trained on 67,600 simulated lab datasets across 13 families of behaviour. It recognises a spectrum, an absorption edge, a thermally activated trend or a saturating isotherm, and proposes the right analysis with one-click Run. Facet also checks your whole project for errors (missing units, too few points, poorly determined parameters, the wrong test), reads graphs (drop a picture of a plot and it traces the data into a table) and runs as an autonomous report agent: it plans the analyses, runs them, checks them and writes the report. Connect your own Claude or GPT key to ask in plain language or attach a figure to have it interpreted and critiqued; the language model plans and explains, and Lattent's validated engine does the maths.
Choose an XY table for curves and spectra, a Column table for comparing groups, or a Multiple-variables table for compositions, process settings and properties. Paste straight from Excel or an instrument export; replicate columns are recognised automatically.
Every analysis produces a figure in a consistent, colour-blind-safe style, exportable as 600-dpi PNG or editable SVG, together with a methods paragraph, a results sentence and a figure legend written from your numbers. Copy them into Word, or download a one-page report for your notebook or supervisor.
Each example opens in Studio with realistic data and the analysis already run. Change the numbers and watch everything update, or replace them with your own.
One-step analyses for the measurements you make every week. Paste your data; the result opens in Studio with its figure and methods text.
Peak fitting, d-spacing, Scherrer and Williamson–Hall.
Open →UV–Vis to optical band gap, direct or indirect, with a confidence interval.
Open →Arrhenius or VFT fits giving Eₐ in kJ/mol and eV.
Open →Langmuir, Freundlich and Sips compared by AICc.
Open →Modulus, 0.2% yield, strength, elongation, toughness.
Open →Raman, FTIR, XPS and chromatography peaks on a fitted baseline.
Open →HPLC, ICP, UV–Vis or ELISA calibration with unknowns and LOD/LOQ.
Open →Any XY data: shape recognition, 35 equations, AICc ranking.
Open →Also available: life-science tools for IC50, enzyme kinetics and qPCR.
Every analysis, table and warning is free, without limits. Pro adds what goes into papers and reports, more cloud storage and the AI chat, for less than a coffee a month.
for everyone, forever
or $4 per month
up to 10 people, one invoice
Prices in US dollars; rupee prices are on the pricing page. Secure checkout by Lemon Squeezy worldwide or Razorpay in India; Lattent never sees card details. Pay by card, PayPal or UPI, or by invoice for institutions; compare plans and payment options. Cancel any time from your account. 14-day refund, no questions. How we keep your data safe.
Lattent is built around the measurements materials and chemistry labs actually make (XRD, spectra, band gaps, isotherms, Arrhenius data, tensile tests), with curve fitting, statistics and publication graphs underneath, plus a shape model, Bayesian optimisation and Facet, an AI lab partner. It runs in any browser with nothing to install and costs a fraction of a desktop licence. It does not yet do everything those programs do, such as 3D plots or two-way ANOVA; those are on the roadmap.
Not unless you choose to. Calculations run in your browser and projects are kept in your browser. If you save a project to the cloud, it is stored in a database where row-level security lets only you, and your lab if you share it, read it. The optional AI chat sends data directly from your browser to the provider you chose, with your own key. Details on the security page.
Yes. Checkout is run by Lemon Squeezy, a PCI-compliant merchant of record that also handles tax. Lattent never sees your card. Your plan is activated by a signed message from Lemon Squeezy to our server, never by your browser.
Every analysis is checked against SciPy and NumPy, the reference scientific Python libraries, on test datasets before each release: parameters, standard errors, p-values, Tukey intervals and Gaussian-process predictions. The fitter is also benchmarked on NIST’s certified nonlinear-regression problems, where it gets 51 of 52 fits right to at least four significant digits. See the validation page.
Open a worked example close to your field, look at what Lattent reports, then replace the numbers with yours. Or create a table, paste your data and press Suggest an analysis.
No. The built-in suggestions need no key. The chat is optional and uses your own Anthropic or OpenAI key, billed by them at their API rates, usually cents per conversation.
Yes. Lemon Squeezy issues a proper invoice with your institution's details; for purchase orders, use the Contact link at the foot of this page.
It proposes new chemical compositions that its models predict to be stable and close to a target band gap, and tells you which are absent from a database of 5.8 million computed structures. That is a shortlist for a DFT calculation or a synthesis, not a discovery in itself: the generator works at the level of composition, not crystal structure, and composition-only predictions have errors of about 0.09 eV/atom for formation energy and 0.3 eV for the band gap. The methods and test results are on the validation page.
No. Lattent is research software and is not validated for GLP, 21 CFR Part 11 or diagnostic use.
No account, no installation, nothing uploaded.