grai
grai is deepsoil's AI soil analysis tool: photograph a sample in a photobox with any smartphone and receive a full particle size distribution curve in seconds, replacing conventional laboratory sieve analysis on site.
What is grai?
grai, short for grading AI, is the flagship product of deepsoil, a spin-off of BOKU, the University of Natural Resources and Life Sciences in Vienna, where the underlying technology was originally developed. It predicts a soil's particle size distribution directly from a photograph taken with an ordinary smartphone, and is positioned as a replacement for sieve analysis, the laboratory procedure that normally ties up equipment and takes days. The method is a deep-learning regression run on the image itself. deepsoil describes automatic scale detection, texture and colour normalisation, detailed PSD curve generation, and cloud-based reporting with historical tracking, and states that the model was trained on thousands of samples. The current model, grai4, outputs eleven PSD classes through a softmax layer. In practice the tool returns a complete PSD curve plus the D10, D30, D60, Cu and Cc indices, and exports to CSV, AGS or a PDF report suitable for engineering documentation. Hardware requirements are deliberately modest: a smartphone and a photobox, a dark chamber that standardises lighting. Free build plans are downloadable, two stackable euroboxes and an LED strip cost around 30 EUR from a third-party retailer, or deepsoil can supply one. The tool needs the phone's pixel density in pixels per millimetre. That value is detected automatically from EXIF data and a calibration database for roughly twenty recognised handsets, falls back to a 70-degree angle-of-view approximation otherwise, and a manual calibration tool is provided on the page. The user manual specifies that images are normalised to a uniform 4.6 pixels per millimetre. There are two ways to consume grai: the web platform, which includes a live drag-and-drop demo, and a REST API, the grai Soil Analysis API v4.0, authenticated with an x-api-key header and documented publicly through Swagger and an OpenAPI 3.1 specification. grai is the first of an announced family: monet for gravimetric water content, lilinet for liquid limit and Atterberg testing, and sonet, which infers particle size from the sound of poured soil, are all listed as coming soon. Since August 2026, grai4 has been integrated with GGU Software's GGU-Connect environment.
What it does
- Predict a soil's full particle size distribution from a single smartphone photograph
- Return the D10, D30, D60, Cu and Cc indices used in geotechnical classification
- Export results as CSV, AGS or a PDF report ready for engineering documentation
- Replace on-site sieve analysis without any laboratory equipment
- Run analyses programmatically through a documented REST API
- Calibrate a phone's pixel density automatically from EXIF data or manually
- Feed image-based soil analysis into the GGU-Connect geotechnical environment
When to use grai / When not to
A quick filter to help you decide if grai is the right fit.
When to use grai
- Geotechnical engineers who need particle size distribution results on site instead of waiting days for a laboratory
- Geotechnical and construction materials testing laboratories looking to cut the cost and turnaround of routine sieve analysis
- Civil engineering and construction site teams running soil investigation at regular intervals across an infrastructure project
- Agronomists and soil scientists who want to read soil composition directly at the farm or field site
- University research groups and lecturers in soil mechanics, who are offered preferential pricing and a data-partnership route to free access
When not to use grai
- Buyers who need contractual guarantees: the publisher currently posts no terms of service, no privacy policy and no data processing agreement
- Teams needing moisture content or Atterberg limits, which are announced as separate future products rather than part of grai
- Anyone hoping to skip laboratory work entirely, since the manual still requires samples to be homogenised and oven-dried before photographing
- Users unwilling to build, buy or obtain a photobox and to calibrate their phone's pixel density
- Organisations that require a documented hosting jurisdiction or a named subprocessor list before uploading project data
How to use grai
A typical end-to-end flow, from setup to results.
- Obtain a photobox: download the free build plans, assemble one from two euroboxes and an LED strip for around 30 EUR, or request one from deepsoil
- Prepare the sample as the manual instructs, homogenising it, oven-drying it and spreading it evenly in the lower tray
- Register for free on the site to try grai4, or request an API key for unlimited access
- Calibrate your phone's pixel density, or leave it on Auto if your handset is one of the recognised models
- Fill the photobox with the soil sample and position the phone over the opening
- Take the photograph through the dedicated aperture
- Upload the image to the grai platform by drag and drop, in JPEG, PNG or WEBP
- Wait a few seconds while the deep-learning regression returns the distribution
- Read the PSD curve together with the D10, D30, D60, Cu and Cc values
- Export the result as CSV, AGS or a PDF report
Pros & Cons
Pros
- Results in seconds where sieve analysis takes days, with deepsoil advertising a three-second analysis
- Usable in the field, with no need to ship samples to a laboratory
- Very low hardware barrier: a smartphone and a photobox costing around 30 EUR, whose plans are given away free
- Prices are published openly, which is unusual in geotechnical software, and a genuine free tier exists
- A publicly documented REST API with Swagger, OpenAPI 3.1 and per-key usage statistics
- Outputs in working formats, with CSV, AGS and PDF reports ready for engineering documentation
- Backed by published research and already integrated into an established geotechnical environment
Cons
- No legal documentation whatsoever: no terms, no privacy policy, no Impressum and no DPA, with thirteen standard legal URLs all returning 404
- No GDPR mention at all, although the publisher is established in Austria, and analytics plus advertising scripts load without any consent banner
- No postal address, no full legal name and no Firmenbuch number are published on the site
- The published pricing contradicts itself: Pay-per-Use costs 49.99 EUR a month for 100 analyses while Premium costs 34.99 EUR a month for unlimited analyses
- Nothing is published about where data is hosted or which subprocessors are involved
- Accuracy depends on calibration, since outside roughly twenty recognised handsets the tool assumes a 70-degree angle of view, an approximation the publisher flags itself
- A one-person team, a single named email address as the only contact, and customer references that are mostly trials and pilots rather than production deployments
Pricing & Plans
A permanent free tier is available: registration on the site is free and gives access to grai4, while an API key is required for unlimited use. The lowest paid entry point is the Premium plan at 34.99 EUR per month. A perpetual licence is also offered at 499 EUR as a one-off purchase, with an upgrade to the newest model priced at 249 EUR.
- 34.99 EUR per month - full API access to grai4
- 11-class PSD analysis
- unlimited analyses
- priority support
- 49.99 EUR per month - 100 analyses per month
- full API access
- priority support
- monthly auto-replenishment
- 499 EUR one-time - unlimited use of the purchased model version
- no recurring fees
- cloud-hosted on deepsoil's infrastructure
- update to the newest model for 249 EUR
- on demand - custom model trained on the customer's data
- optimised for their soil conditions
- dedicated deployment
- ongoing support
- tailored pricing - on-site or online sessions covering photobox use
- image capture best practice
- result interpretation and workflow integration
- free - grai4 access in exchange for contributed data
- with the access period based on data quality
- plus research publication collaboration and co-development
Data, GDPR & hosting
A consolidated view of how grai handles your data.
GDPR overview
There is no GDPR mention of any kind. Searches across the site's HTML and its JavaScript return zero occurrences of GDPR, DSGVO or Datenschutz. deepsoil publishes no privacy policy, names no data controller, designates neither a data protection officer nor an Article 27 representative, and documents no data subject rights, retention periods, subprocessors or non-EU transfers. This matters because the publisher is established in Austria and therefore falls squarely within the scope of the regulation. Two further points are worth noting: the site carries no Impressum at all, although publishing legal notices is a national requirement in Austria, and it loads Google Analytics and Google AdSense on the homepage without any cookie banner or consent mechanism. This absence of statements is not a finding of non-compliance; it simply means nothing is documented for a buyer to assess.
Who owns the data?
No terms and conditions are published, so no document states who owns the soil images uploaded to grai or the particle size results they produce. Thirteen of the usual legal URLs on the domain return 404, and the site contains no ownership clause of any kind. The only contractual language visible anywhere is the perpetual licence, which grants unlimited use of the purchased model version, a licence over the model rather than a statement about customer data. The optional Data Partnership offer implies that any transfer of data to deepsoil is voluntary and negotiated case by case. Ownership therefore remains formally undefined.
Reuse rights
Nothing on the site defines what a customer may do with the outputs, nor what deepsoil may do with uploaded images, because no terms and conditions exist. What can be observed is that images leave the user's device and are processed by a remote API, and that this API states it keeps usage records, namely request counts, success rates, response times and per-endpoint breakdowns, stored against a hashed API key. Two commercial offers do involve customer data explicitly: Fine-Tuning trains a custom model on the client's own data, and Data Partnership exchanges free grai4 access for contributed data. Outside those two cases, reuse rights are simply not addressed.
Data retention & training
Hosting summary
deepsoil publishes no trust, security or privacy page, so no hosting jurisdiction is declared anywhere for customer data. What follows is observation of the infrastructure, not a commitment by the publisher. The deepsoil.at website resolves to 88.198.195.20, an address belonging to Hetzner Online GmbH (AS24940) in Nuremberg, Germany. The grai API is served from a Hugging Face Space published under the account soranz84, whose public metadata carries a region:us tag. The perpetual licence description mentions only that the purchased model version is cloud-hosted on the publisher's own infrastructure, without saying where. Taken together, these signals suggest that website traffic terminates in Germany while API processing may occur outside the European Union, but none of this is stated by deepsoil and none of it constitutes a contractual guarantee. Buyers who need a defined hosting jurisdiction should obtain it in writing before uploading project data.
Things to keep in mind
Risks and trade-offs to weigh before adopting grai.
- No published terms, privacy policy or DPA means you upload project data under no stated rules, so check with the publisher before sending client material
- The publisher does not say where data is hosted; the API is served from a Hugging Face Space tagged for a US region, which is an infrastructure observation rather than a commitment
- Google Analytics and Google AdSense load on the homepage with no cookie banner or consent step, which visitors in the European Union should be aware of
- Advertised figures such as three-second analysis and over 90% cost savings are publisher claims, published without accuracy metrics on the site
- Treating an AI estimate as equivalent to a certified sieve test is the main misuse risk; results feeding structural or safety decisions should still be cross-checked against laboratory methods
- Outside the roughly twenty calibrated handsets a 70-degree angle-of-view assumption is applied, so switching phones can silently change results
- The publisher is a one-person operation reachable at a single email address, a continuity risk for anyone embedding grai in a long-running workflow, and no geographic restriction is announced anywhere on the site
Setup & Integrations
Technical difficulty
Software setup is easy: register free on the site and drag an image into the demo. The friction is physical. You need a photobox, either built from the free PDF plans, assembled from two euroboxes and an LED strip bought elsewhere, or ordered from deepsoil, and you must know your phone's pixel density, which is automatic on roughly twenty recognised handsets and manual otherwise. The manual also requires samples to be homogenised and oven-dried. API use is standard multipart HTTP with a key header. deepsoil sells a training service, which suggests it does not consider onboarding trivial.
Deployment
Integrations
Behind grai
Social
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement grai.
Frequently asked questions
What exactly does grai measure?
What equipment do I need?
How long does an analysis take?
In what formats can I export the results?
Is there an API?
Is there a free option?
What does the cheapest paid plan cost?
Will it work with my phone?
Does the publisher provide terms of service or a privacy policy?
What else is deepsoil building?
Should you pick grai?
grai is a narrow tool that does one thing and does it cheaply: it turns a smartphone photograph of a soil sample into a particle size distribution curve, in seconds rather than the days a laboratory sieve analysis takes. For a geotechnical engineer, a materials testing laboratory or a site team running repeated soil investigation, that is a genuinely useful trade, and the entry cost is unusually low, with free build plans for the photobox, a free registered tier, and a paid entry point at 34.99 EUR per month. The technical case is credible. deepsoil publishes four papers behind the model, comes out of BOKU in Vienna, and has already shipped one real product integration with GGU-Connect. Several geotechnical firms and testing laboratories are shown testing or piloting it, and six universities are listed as research collaborators, though this should be read for what it is: early traction built on trials and pilots rather than mass deployment. The weakness is not technical but contractual. deepsoil publishes no terms of service, no privacy policy, no Impressum and no data processing agreement, gives no postal address, and says nothing about where data is hosted. That is unusual for a publisher established in the European Union, and a real obstacle for anyone buying through a procurement process. The pricing grid also needs attention, since the metered plan is currently dearer than the unlimited one. Worth trying if you want to test image-based granulometry at low cost. Ask for the legal and hosting details in writing before sending it client project data.
- Choosing a selection results in a full page refresh.
- Opens in a new window.