Neptune
Neptune is a dedicated central ML metadata store that logs and queries model-building metadata, providing a web app for visualization, comparison, monitoring, and collaboration across machine learning experiments.
Publisher review
Neptune is a dedicated central ML metadata store that logs and queries model-building metadata, providing a web app for visualization, comparison, monitoring, and collaboration across machine learning experiments. It is designed for data scientists and ML engineers who need a centralized hub to organize runs, compare results, and share findings with their teams. The platform offers a flexible API for logging and fetching experimentation metadata, with integration into various frameworks in the ML ecosystem, making it suitable for both no-code and low-code environments. Originally developed as an internal tool at deepsense.ai in 2016, Neptune spun out as a standalone product in 2018 and has since been acquired by OpenAI, with public services winding down by March 2026.
Neptune works by providing a central metadata store where users log metrics, parameters, artifacts, and model versions from their ML experiments. Its API allows for flexible logging and querying, supporting integrations with popular frameworks like TensorFlow, PyTorch, and scikit-learn. The web app offers a user-friendly drag-and-drop interface for building dashboards and comparing runs, enabling quick application development without extensive coding. The platform's collaboration features allow teams to share and discuss results in real-time, with a Starter plan offering 3 active projects and up to 10 users for $150 per month. Neptune's metadata store can handle large volumes of experiment data, with support for custom metadata structures and automated logging through its client library.
In the market, Neptune competes directly with MLflow, Weights & Biases (WandB), and ZenML. Unlike MLflow, which offers a standalone open-source solution, Neptune is proprietary SaaS with an Apache 2.0-licensed client but no standalone OSS option. Compared to WandB, Neptune provides a more focused experiment tracking experience but lacks features like dataset versioning and user access management that WandB offers. ZenML treats ML workflows as pipelines, whereas Neptune centers on experiment runs, making it less suited for end-to-end pipeline orchestration. Neptune's acquisition by OpenAI signals strong adoption in large-scale model training but also raises concerns about long-term availability, as the service will be permanently discontinued after March 2026.
The honest trade-offs with Neptune include its lack of a standalone open-source solution, which limits flexibility for teams wanting full control over their infrastructure. It also lacks certain features like dataset versioning and user access management, which are available in competitors like WandB. The most significant trade-off is that Neptune's public service will be discontinued by March 2026, making it unsuitable for long-term projects without a migration plan. While its drag-and-drop interface and quick application development are strengths, the platform's reliance on a proprietary SaaS model and impending shutdown mean users must plan for alternatives.
How it works
-
Central ML metadata store
Logs and queries model-building metadata, providing a single hub for organizing ML runs and comparing results.
-
Web app for visualization
Offers drag-and-drop interface for building dashboards, comparing runs, and monitoring experiments in real-time.
-
Flexible API for logging
API allows logging and fetching of metrics, parameters, artifacts, and model versions from any ML workflow.
-
Framework integrations
Integrates with TensorFlow, PyTorch, scikit-learn, and other ML ecosystem frameworks for automated metadata logging.
-
Collaboration features
Enables teams to share results, compare experiments, and collaborate through the web app with real-time updates.
-
No-code and low-code support
Provides drag-and-drop app development for quick prototyping, alongside API-based logging for advanced users.
-
Starter plan with 3 projects
Starter plan includes 3 active projects and up to 10 users for $150 per month, suitable for small teams.
Strengths and trade-offs
Strengths
- Central hub to organize ML runs and compare results, with a user-friendly drag-and-drop interface for app development.
- Quick application development is enabled through no-code and low-code environments, reducing time to insight.
- Flexible API for logging and fetching metadata supports custom workflows and integration with diverse ML frameworks.
- Collaboration features allow teams to share and discuss experiment results in real-time, improving productivity.
Trade-offs
- Does not offer a standalone open-source solution, limiting flexibility for teams wanting full infrastructure control.
- Lacks certain features like dataset versioning and user access management, which are available in competitors like WandB.
- Services will be permanently discontinued after March 2026 due to acquisition by OpenAI, requiring migration planning.
- Proprietary SaaS model with no self-hosted option creates dependency on vendor infrastructure and pricing.
Pricing context
Starter plan includes 3 active projects and up to 10 users for $150 per month; no public pricing for higher tiers.
Getting started with Neptune
-
Sign up for Neptune
Go to the Neptune website and create an account. Choose the Starter plan for 3 active projects and up to 10 users at $150 per month, or contact sales for higher tiers. Verify your email to activate the account.
-
Install the Neptune client
Install the Neptune client library in your Python environment using pip: pip install neptune-client. This library provides the API for logging and fetching metadata from your ML experiments.
-
Connect to your project
Log in to the Neptune web app and create a new project. Copy your project API token from the settings. In your code, initialize a Neptune run with neptune.init_run(project='your-project', api_token='your-token') to connect.
-
Log experiment metadata
In your ML script, log parameters, metrics, and artifacts using the Neptune run object. For example, run['parameters'] = {'lr': 0.01} and run['metrics/accuracy'].log(0.95). This stores data in the central metadata store for later comparison.
-
Visualize and compare runs
Open the Neptune web app and navigate to your project. Use the drag-and-drop interface to build dashboards, compare runs side by side, and monitor real-time updates. Share results with your team via the collaboration features.
Frequently Asked Questions
What is Neptune in machine learning?
Neptune is a central ML metadata store that logs and queries model-building metadata. It provides a web app for visualizing, comparing, monitoring, and collaborating on machine learning experiments. It is designed for data scientists and ML engineers to organize runs and share findings.
How much does Neptune cost?
Neptune's Starter plan costs $150 per month and includes 3 active projects and up to 10 users. Pricing for higher tiers is not publicly listed. The platform is a proprietary SaaS with no standalone open-source option, and services will end by March 2026.
How does Neptune compare to MLflow and Weights & Biases?
Neptune is a proprietary SaaS, unlike MLflow which offers a standalone open-source solution. Compared to Weights & Biases, Neptune provides focused experiment tracking but lacks dataset versioning and user access management. Neptune centers on experiment runs, not end-to-end pipelines like ZenML.
What are the main features of Neptune?
Neptune offers a central ML metadata store, a web app with drag-and-drop dashboards, a flexible API for logging metrics and artifacts, integrations with TensorFlow and PyTorch, and collaboration tools for real-time sharing. It supports no-code and low-code environments for quick prototyping.
Is Neptune being discontinued?
Yes, Neptune was acquired by OpenAI and its public services will be permanently discontinued after March 2026. Users need to plan a migration to alternative tools. This makes Neptune unsuitable for long-term projects without a migration strategy.
What are the strengths and weaknesses of Neptune?
Strengths include a user-friendly drag-and-drop interface, flexible API, and real-time collaboration. Weaknesses are no standalone open-source option, missing dataset versioning and user access management, and the impending shutdown by March 2026, requiring migration planning.
Alternatives
How Neptune compares
Direct head-to-head against 3 competitors. Picked by 7wData.
Neptune
- Pricing
- Starter plan includes 3 active projects and up to 10 users for $150 per month; no public pricing for higher tiers.
- Target
- Neptune is a dedicated central ML metadata store that logs and queries model-building metadata, providing a web app for visualization, comparison, monitoring, and collaboration across
- Strength
- Central hub to organize ML runs and compare results, with a user-friendly drag-and-drop interface for app development.
- Watch for
- Does not offer a standalone open-source solution, limiting flexibility for teams wanting full infrastructure control.
PuppyGraph
- Pricing
- Forever-free Developer Edition, enterprise pricing custom
- Target
- Teams needing zero-ETL graph queries on existing data lakes/warehouses
- Deployment
- Docker, AWS/GCP
- Strength
- Petabyte-scale analytics without data migration
- Watch for
- Newer entrant vs established graph DBs
Neo4j
- Pricing
- $0.58/hour for Aura Professional (cloud), self-hosted from $9,995/server/year
- Target
- Enterprises requiring native graph processing
- Deployment
- Cloud, on-prem, hybrid
- Strength
- Industry-standard Cypher query language
- Watch for
- Cloud costs escalate with large datasets
TigerGraph
- Pricing
- Custom/Contact sales
- Target
- Large-scale parallel graph analytics
- Deployment
- Cloud, on-prem
- Strength
- Deep-link OLAP capabilities
- Watch for
- Enterprise pricing requires negotiation
User reviews
No user reviews yet. Be the first to write one.
Sources
Reporting on this tool draws on these publicly available sources.