We Raised $8M Series A to Continue Building Experiment Tracking and Model Registry That “Just Works”
Unlimited users on all paid plans. Start for free, then scale as needed
“It is simply natural to pay for usage, not users.”
What are the monitoring hours?
Monitoring hours is our way of measuring the time spent actually logging metadata in Neptune via the API.
When you log metadata to a run we create a new logging session. If any additional data is tracked in the next 10 minutes, the session time is increased by the time elapsed between those two events and rounded up to the nearest second – otherwise the session ends. If the session has only one event the session length is 1 second.
Your monitoring hours for this run is the sum of all logging sessions.
- Your run took 100 minutes.
- You logged metadata at minutes 12, 18, 40, 60, a thousand times between minute 70 and 76, and 80.
- You pay for 6 monitoring minutes in logging session 1, 1 second for logging session 2 (single log) in minute 40, and 20 monitoring minutes in logging session 3, for a total of 26 minutes and 1 second.
Monitoring hours of different runs are calculated independently. The monitoring hours of all runs across all your projects are summed up to a total monitoring hours amount for your workspace.
How do you guesstimate your monthly usage?
Take a look at a few examples below. They should help you estimate your monthly usage.
Project types: time series prediction
Neptune use cases: experiment tracking and reproducibility
Runs per month (in average): 100
Monitoring hours usage: 300 h/monthly
Storage usage: 300 GB
How much they pay? $170/month
Project types: object detection
Neptune use cases: experiment tracking, data versioning, and monitoring model training
Runs per month (in average): 600
Monitoring hours usage: 800 h/monthly
Storage usage: 100 GB
How much they pay? $150/month
Project types: NLP and reinforcement learning
Neptune use cases: experiment tracking, monitoring model training, model versioning, and integrated Neptune in the retraining pipelines
Runs per month (in average): 2000
Monitoring hours usage: 4 000 h/monthly
Storage usage: 500 GB
How much they pay? $640/month
Project types: Computer vision and NLP
Neptune use cases: experiment tracking, model registry, and integrated Neptune in CI/CD pipelines
Runs per month (in average): 15000
Monitoring hours usage: 20 000 h/monthly
Storage usage: 1 TB
How much they pay? Custom deal
If you are training your models in the cloud, look at your monthly usage there and use it as an upper bound for your Neptune usage. It will be less than that as we count monitoring hours only when you log metadata.
“The more I used Neptune, the more I felt that I would rather pay for a hosted solution than have to maintain the infrastructure myself…”
“Neptune made sense to us due to its pay-per-use or usage-based pricing. Now when we are doing active experiments then we can scale up and when we’re busy integrating all our models for a few months then we scale down again.”
“Can I use Neptune for free?” And more questions
Payments and subscription
Can I use Neptune for free?
Yes! You can use Neptune for free for work, research, and personal projects. On the Individual plan you receive a monthly quota of 200 monitoring hours (unused hours pass to the next month and only expire after a year) and a storage limit of 100 GB.
If you are working in academia check Neptune program for researchers.
Many tools charge per user. Why do you have usage-based pricing?
We believe value-based pricing is fair. What you pay should correlate with the value you are getting from the tool. It’s that simple. We use a heuristic based on the run monitoring time recorded the amount of data stored to estimate that value.
Secondly, this approach fits any model-building life cycle. Many data science teams do not train models 24/7 but have a few weeks of heavy model training every 3-4 months. With value-based pricing, you only pay when you are actually using the tool.
Lastly, You should be able to share your ML metadata in your organization with Data scientists, ML engineers, DevOps engineers, and managers when you need to. With user-based pricing, you’d have to pay extra. With value-based pricing, you can invite people whenever you want.
Is it really $150/monthly for the whole team?
Yes! You can invite your entire ML team, people from the product division, CEO, and your company dog. It will still be the same price.
The whole team really means the whole team.
Only if they actively log metadata to Neptune they will be using the monthly quota. Browsing UI doesn’t count.
How do I buy monitoring hours?
Each month, depending on the plan you’re on, a specified amount of monitoring hours is added to your account. Additional monitoring hours are purchased in blocks of 100. You can buy them anytime – just head to the Subscription tab of your workspace to top up your account.
Will I be charged automatically when my monthly quota is over?
No, we won’t charge you automatically. You need to top up your account and decide on how many additional monitoring hours or storage you’d like to purchase.
Do you have any Startup discounts?
If you are a Startup you can get a one-time bonus of 18000 monitoring hours added for free to your Team or Organization plan.
Contact us at firstname.lastname@example.org to apply for the Startup program.
Can I get free access for Academia or Kaggle or Non-profit?
We offer a free Academia license for Team plan with an increased monthly quota for research, educational, non-profit organizations, and Kaggle teams.
Set up a team workspace and claim the Academia licence in the process.
We will verify teams’ usage with the Academia license and deactivate workspaces that we believe are breaking the license rules. Any questions? Reach out to email@example.com – Researchers program advocate at Neptune.
What are the monitoring hours?
The monitoring hours refer to the amount of time you use Neptune to track your training runs or monitor models on production.
Browsing and exploring your data through the web application does not use any monitoring hours. Neither does fetching data programatically in the read-only connection mode. The monitoring hours are only counted when you send metadata to Neptune, typically while tracking runs using Neptune Python client library.
You can always check your current balance of monitoring hours on the Subscription page of your workspace. Neptune implements per-minute billing.
What happens when I run out of monitoring hours?
When you run out of monitoring hours your workspace will enter a read-only mode until you return to a positive monitoring hours balance. This will happen when you top up your account or when you receive your new monthly package of monitoring hours.
While in read-only mode you have full access to already tracked metadata. You can browse or analyze them through the web application and fetch programmatically with the read-only connection mode. However, you won’t be able to track new runs or upload new metadata to the existing ones.
If you run out of monitoring hours while logging new metadata (e.g., training models), the client library will automatically enter offline mode. All metadata tracked after that will be stored locally and can be later be uploaded with the Neptune Command Line Interface. Don’t worry – when you will be running out of the monitoring hours, we will let you know in advance so that it doesn’t catch you as a surprise.
How does read-only mode work?
Your workspace will switch to the read-only mode if you run out of monitoring hours, exceed your storage quota, or in case of Team/Scale plans if Neptune has not been able to charge the monthly subscription fee for the last 14 days.
While in read-only mode you have full access to your already tracked metadata. You can browse or analyze them through the web application and fetch programmatically with the read-only connection mode. However, you won’t be able to track new runs or upload new metadata to the existing ones. You can also delete old data either through the web application or programmatically through the Python API.
Do monitoring hours expire?
Unused monitoring hours only expire one year after you purchase or receive them. This includes the monitoring hours you receive as part of your plan each month. When the account subscription is canceled unused monitoring hours will be forfeited.
Do monitoring hours include using the UI to view experiments?
No! Browsing and exploring your data through the web application does not use any monitoring hours. Neither does fetching data programatically in the read-only connection mode.
The monitoring hours are only counted when you send metadata to Neptune, typically while tracking runs using Neptune Python client library.
How many monitoring hours do I need?
As a rule of thumb the monitoring time is directly proportional to the time you spend on training the models.
If you are using a public cloud provider for the training infrastructure you can simply check how much time you spent on training the models. To monitor the whole training process you will need a similar amount of monitoring hours. If you are using your own infrastructure and don’t have a thorough monitoring setup try to estimate how many models you train every month/quarter. If every training job takes a similar amount of time simply multiply the numbers, if not you will need to estimate what’s the average training time.
Remember that different teams have different patterns of model training even within the same company. Some teams train new models on a regular basis every week, some teams have a few weeks of heavy model training every 3-4 months. To get the best estimate, find out your pattern and calculate an average over a longer period.
What happens when I exceed my storage quota?
Each workspace starts with 100GB of storage. When you exceed your storage quota, your workspace will enter a read-only mode. To regain full access, you can either increase your storage quota or delete some old data.
While in read-only mode, you have full access to already tracked metadata. You can browse or analyze them through the web application and fetch programmatically with the read-only connection mode. However, you won’t be able to track new runs or upload new metadata to the existing ones. You can also delete old data either through the web application or programmatically through the Python API.
If you exceed your storage quota while logging new metadata (e.g. while training models) the client library will automatically enter offline mode. All metadata tracked after that will be stored locally and can later be uploaded with the Neptune Command Line Interface. Don’t worry – when you’re close to exceeding the storage quota we will let you know in advance so that it doesn’t catch you by surprise.
If I store my tracked data on an external server like AWS S3 or GCS does it count against the storage quota?
Absolutely no. Only the size of the metadata that you actually store in Neptune counts against your storage quota.
For example, if you have a 2 GB dataset of images and track it using the Artifacts functionality, a list of the files, their size, and hash sum is stored in Neptune, taking up a few MBs of storage at most.
Features and benefits
What are service tokens?
Service token is a dedicated security token for automated MLOps pipelines or CI/CD.
What dedicated support can I get?
A dedicated support person will get you help when you need it. That includes helping you with the setup, answering your product-related questions, debugging issues, or setting up an onboarding session for your new teammates.
Where is SaaS hosted?
Currently we are hosted on the Google Cloud Platform in Europe.
Can I deploy Neptune on-prem?
Yes! You can deploy Neptune on your own servers. At the top of the page check out the “Your Server” tab.