Graviti
Graviti is a SaaS data platform built for machine learning teams to host, version, filter and visualize unstructured datasets such as images, video, lidar and audio, and to automate the pipelines that process them.
What is Graviti?
Graviti Data Platform is a SaaS product for managing the unstructured data that feeds machine learning and business analytics. Its pitch, Unleash the Power of Unstructured Data, rests on an argument the company sets out on its about page: two decades of software have been built around structured data, while video, images and audio files pile up faster than ever with no centralized platform to hold them.
The answer is to drop the folder tree. The unit of management is the dataset, holding raw data, annotations and metadata together under standardized annotation formats - Classification, Box, Polygon, Polyline, Cuboid, 2D Box, 2D Polygon, 2D Polyline and 3D Box. It takes images, video and lidar in .pnts, .pcd, .bin, .las and .ply, along with time-series and fusion datasets.
Four services sit around that core. Version control behaves like Git: commits, branches, tags and drafts, with commits read-only and a single writer per draft, and new datasets built by split or merge without consuming extra storage. Visualization runs online with no download and no code, and its widgets embed into an external page with one line. Collaboration uses role-based access control at dataset-per-user level, with link sharing and access approval. Workflow automation takes pipelines written in YAML, runs them in containers with workers spawned on demand, and triggers them from cron, a new commit or a new branch.
Developers reach the same functions through a Python SDK, a CLI and an OpenAPI, documented as TensorBay. Two applications ship with the platform, GroundTruth for collaborative annotation and Sextant, alongside a marketplace for third-party tools. Graviti also hosts open datasets free of charge, its catalogue including nuScenes, BDD100K and MNIST.
The company behind it is GRAVITI TECHNOLOGIES INC, founded by Edward Cui, formerly a machine learning specialist in Uber's Advanced Technologies Group. The interface is offered in English and Chinese. One caveat outweighs any feature: the marketing site answers normally, but the application itself, at gas.graviti.com and account.graviti.com, no longer resolves in DNS.
What it does
- Host raw data, annotations and metadata together in a single versioned dataset
- Version datasets with commits, branches, tags and drafts, and diff any two versions
- Query and filter on metadata and semantic data, then build a new dataset by filter or merge with no copying
- Visualize images, video and lidar online, with label and attribute distribution charts
- Automate processing pipelines triggered by cron, a new commit or a new branch
- Share datasets across a team with role-based access control and access requests
- Drive the whole platform from a Python SDK, a CLI or an OpenAPI
When to use Graviti / When not to
A quick filter to help you decide if Graviti is the right fit.
When to use Graviti
- Machine learning engineers managing large image, video and lidar datasets across model iterations
- Autonomous driving and ADAS teams working with 3D boxes, point clouds and sensor-fusion datasets
- Data and MLOps engineers who want dataset versioning and scheduled processing without building the infrastructure in-house
- Research groups and universities looking for free hosting and a catalogue of open datasets such as nuScenes, BDD100K and MNIST
- Developer-led teams that would rather drive a data platform through a Python SDK, a CLI and an OpenAPI than through a user interface
When not to use Graviti
- Teams working exclusively with structured, tabular data - the platform is positioned as the opposite of that
- Anyone who needs to sign up and start today: the application subdomains no longer resolve in DNS
- Buyers who require a readable privacy policy or contract before committing, since the legal pages are offline
- Small teams and individuals: the first paid tier starts at USD 200 per month
- Users looking for a standalone annotation tool - GroundTruth is a module inside the platform, not a separate product
How to use Graviti
A typical end-to-end flow, from setup to results.
- Sign up at account.graviti.com - note that this address no longer resolves, so the steps below describe the documented path rather than one you can follow today
- Create a dataset from the web interface, from a local upload, by migrating from a cloud, or by authenticating an existing cloud storage account
- Upload raw data, annotations and metadata into that single dataset
- Commit a version whenever you want; the complete record of iterations is kept
- Inspect what you have with the online visualization widgets, without downloading anything
- Add custom columns such as labels or evaluation results - they are indexed and can be queried through filters
- Filter on metadata and semantic data, then create a new dataset by filter or by merge
- Open a branch or a draft so several people can work without disturbing each other
- Configure a workflow in YAML and choose its trigger: cron, a new commit or a new branch
- For automation and integration, switch to the Python SDK, the CLI or the OpenAPI
Pros & Cons
Pros
- Covers hosting, versioning, visualization, curation, collaboration and automation in a single product
- Incremental versioning with no duplicated storage, which is the central argument of the solution pages
- Unlimited seats on both paid tiers
- A permanent free tier with 100 GB of storage, 50,000 rows and two hours of XS compute
- Free hosting for open datasets, with a reference catalogue including nuScenes, BDD100K and MNIST
- Usage pricing published in full, down to the price of a terabyte of transfer
- ISO/IEC 27001:2013 and ISO 9001:2015 certifications displayed, and lidar formats and fusion datasets supported, which is uncommon on this segment
Cons
- The application is unreachable: gas.graviti.com and account.graviti.com have no DNS A record
- The privacy policy, terms and cookie policy are offline, so nothing the vendor commits to can be read
- No postal address and no email address are published; the only direct contact is a US toll-free number
- The site is frozen - a 2021 footer copyright, a sitemap last generated in October 2023, and dead careers and news pages
- A high entry ticket, with the first paid tier at USD 200 per month
- Nothing published on data retention, subprocessors or whether models are trained on customer data
- No mobile app, no browser extension, and the advertised 14-day trial link no longer responds
Pricing & Plans
Graviti offers a permanent free plan. The Starter tier costs nothing and includes 100 GB of private cloud storage, 50,000 rows and two hours of XS compute. The lowest paid entry point is the Standard plan, advertised from USD 200 per month with unlimited seats, on top of which usage is billed at USD 40 per terabyte per month for storage, USD 0.10 per 1,000 rows per month and USD 1 per hour of XS compute. The Premium plan starts at USD 800 per month. Billing is monthly, and custom and on-premise arrangements are available on request.
- 100 GB of private cloud storage
- 50
- 000 rows
- two hours of XS compute
- 5 dataset containers
- a 30-day activity log
- community content only and no third-party applications
- unlimited dataset containers
- a 180-day activity log
- storage at USD 40 per TB per month
- USD 0.10 per 1
- 000 rows per month
- XS compute at USD 1 per hour
- a 14-day free trial
- technical support and product onboarding
- unlimited dataset containers
- a one-year activity log
- storage at USD 40 per TB per month
- USD 0.15 per 1
- 000 rows per month
- XS compute at USD 1 per hour
- a prepaid discount and customer success
- priced on request through Contact Sales for larger sizes
- available on request for teams that need full control over their data
- 20 GB free
- then USD 0 per TB within the same region
- USD 20 per TB to a different region and USD 90 per TB outbound
Data, GDPR & hosting
A consolidated view of how Graviti handles your data.
GDPR overview
GDPR is mentioned exactly once across the whole site. The advantages page, in a block headed Regulatory Compliance in Data Management and Data Security, states that Graviti complies with China's Export Control Law and with EU GDPR requirements on cross-border data transmission. That is a narrow claim: it covers transfers, not the regulation as a whole. The same block cites role assignment, backup, an audit log and a sandbox system, and the data hosting page adds ISO/IEC 27001:2013 and ISO 9001:2015 certifications. Beyond that, nothing. No Article 27 representative is named, no data protection officer is designated, no GDPR contact address is published, and the privacy policy that would let anyone judge the real scope of the claim sits on a subdomain that no longer resolves.
Who owns the data?
The terms and conditions cannot be consulted. Graviti publishes them on the account.graviti.com subdomain, which no longer resolves, so there is no contractual statement about who owns the data or what the vendor may do with it. What the live site shows is indirect only. Graviti holds ISO/IEC 27001:2013 and ISO 9001:2015 certifications, it lets customers authenticate their own existing cloud storage instead of uploading copies, and it offers an on-premise plan described as giving full control over your data. Those arrangements suggest the customer keeps their datasets, but none of them is a commitment. Ownership should be confirmed with Graviti directly before anything sensitive is uploaded.
Reuse rights
No usable answer exists in the public documents, because the terms and conditions sit on a subdomain that no longer resolves. Nothing on the live site says whether a customer may reuse, redistribute or commercially exploit the datasets and workflow outputs the platform produces, and nothing says what Graviti itself may do with uploaded material. Observable behaviour is limited to product mechanics: storage, indexing, filtering, versioning, visualization, pre-annotation and auto-labeling, executed in containers with workers spawned on demand. No mention of training models on customer data appears anywhere in the collected pages. The reuse position therefore has to be settled with the vendor rather than read off the site.
Data retention & training
Hosting summary
Graviti does not publish where customer data is hosted. No country and no region is named anywhere on the site. The pricing page reasons in terms of the same region, a different region and transfers away, charging USD 0, USD 20 and USD 90 per terabyte respectively, but never says which regions those are. Two arrangements reduce the question: customers can authenticate their own existing cloud storage instead of uploading copies, and an on-premise plan is available on request for teams that need full control. On security, Graviti displays ISO/IEC 27001:2013 and ISO 9001:2015 certifications and cites a sandbox system, role assignment, backup and an audit log. The advantages page claims compliance with China's Export Control Law and with EU GDPR requirements on cross-border transmission, which implies operations spanning both jurisdictions without stating where anything is stored. As a technical note concerning the marketing site only and not customer data, www.graviti.com is served by an AWS load balancer in us-west-2, at 54.203.139.165 in Boardman, Oregon. The privacy policy that would settle the question is offline.
Things to keep in mind
Risks and trade-offs to weigh before adopting Graviti.
- The platform itself is unreachable: gas.graviti.com and account.graviti.com point at deleted AWS load balancers, checked against two public resolvers on 24 August 2026
- Uploading proprietary datasets to a vendor whose privacy policy, terms and cookie policy are all offline means accepting terms that nobody can read
- No retention rule is published; the only duration on the site is the activity log, kept 30 days to one year by tier, which is a product feature and not a personal data retention policy
- No postal address and no email address, so a single toll-free number is the entire escalation path if something goes wrong with your data
- The free tier and the open dataset catalogue make it tempting to move research data onto a platform whose continuity is unproven
- Usage billing on storage, rows, compute and transfer can climb well beyond the advertised USD 200 floor with no obvious ceiling
- Outbound transfer is priced at USD 90 per terabyte, so leaving becomes expensive once large datasets are in - a lock-in worth pricing before committing
Setup & Integrations
Technical difficulty
Moderate, and squarely aimed at engineers. The web interface lets you create a dataset, upload, commit and visualize without writing code, but the vocabulary is Git's - commits, branches, drafts, diffs - and familiarity is assumed. Anything beyond the basics means the Python SDK, the CLI, the OpenAPI or YAML workflow files, with containers and environment parameters to configure. Connecting existing cloud storage adds an authentication step on the provider side. Graviti addresses algorithm and ML engineering teams, not general users. One practical obstacle today: the sign-up path at account.graviti.com cannot be reached.
Deployment
Integrations
Supported languages
Behind Graviti
Social
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
What does Graviti actually do?
Which data types and formats does it handle?
Is there a free plan?
How much does the first paid tier cost?
Is there an API?
How does the version control work?
What security certifications does Graviti hold?
Can data stay on my own infrastructure?
Can I use the service today?
Should you pick Graviti?
Graviti built something genuinely coherent on a narrow segment. Pulling raw data, annotations and metadata into a single versioned dataset, then wrapping it in Git-like commits, online visualization of lidar and video, role-based sharing and YAML-driven pipelines, is a complete answer to a real problem - the one its own advantages page quantifies when it reports that algorithm engineers spend sixty percent of their time cleaning and managing data. The pricing is unusually transparent for enterprise infrastructure, published down to the cost of a terabyte of transfer, and the customer names in autonomous driving carry weight.
What has to be said just as plainly is that the operational signals all point the other way. The marketing site answers, but the product does not: gas.graviti.com and account.graviti.com resolve to AWS load balancers that no longer exist, which puts sign-up, login, the dataset workspace, the open dataset catalogue and the application marketplace out of reach. The privacy policy, the terms and the cookie policy live on that same dead subdomain, so a prospective customer cannot read a single contractual commitment. The footer copyright still reads 2021, the sitemap was last generated in October 2023, and the careers and news pages are gone.
Treat this entry as a record of what Graviti offered rather than as a buying guide. If the platform still matters to you, and for a team drowning in unversioned lidar and video it might, the sensible next step is to call the toll-free number the company still publishes and ask what remains in service. Nothing on the site should be relied on as a current commercial offer, and nothing about data handling can be verified while the legal pages stay offline.
- Choosing a selection results in a full page refresh.
- Opens in a new window.