Ways to build
| Concept | What it is |
|---|---|
| A self-serve interactive development environment using the IDE of your choice, designed to feel like a “laptop in the cloud.” Intended for exploratory analysis and interactive development. | |
| A batch run of a script or command, fully reproducible, for training, simulation, and data processing. Can be run manually, via APIs, or on a schedule. | |
| Useful primarily when developing machine learning models, Experiments provide a way to track, organize, and compare different model configurations you’re testing. | |
| Multi-step computation graphs. A good fit for data engineering tasks or MLOps tasks. | |
| Frameworks for running highly parallelizable workloads, including Spark, Ray, Dask, and Slurm. Domino can attach clusters to Workspaces, Jobs, and other workloads and manages the necessary infrastructure, including autoscaling. |
Use data
| Concept | What it is |
|---|---|
| Versioned filesystem storage for large data, shared across Projects, with strong features for reproducibility. | |
| Like Datasets but backed by NetApp ONTAP storage, with near-instant snapshots regardless of size. Use them wherever your organization runs ONTAP. | |
| Connectors to external databases, data warehouses, data lakes, and object stores such as Amazon S3. | |
| Network-attached storage, such as network file system (NFS) shares or Windows shares, mounted into the workloads you run. |
Productize what you build
| Concept | What it is |
|---|---|
| An HTTP API wrapped around your code, meant for real-time inference tasks such as classification and scoring. | |
| Interactive web applications built with frameworks such as Streamlit, Dash, Shiny, Flask, or any other web framework. Domino handles routing, authentication, and scaling. | |
| Programs that use a large language model (LLM) to plan and act across multiple steps. Domino hosts and serves them alongside Apps and traces them for evaluation and monitoring. | |
| Self-serve web forms that run a predefined Job with the parameters a less technical user fills in. Useful for turning scripts into simple web interfaces. |
Govern
| Concept | What it is |
|---|---|
| A router that lets you direct LLM calls to any underlying model, with usage tracking, cost controls, and guardrails. Contact your Domino field team for access. | |
| Policies define the stages, required evidence, and required approvals that a work product, such as an App or model, must go through as it’s built. Used to enforce standards for internal workflows and compliance policies. | |
| Manage and monitor infrastructure spend. | |
| A record of user activity for compliance verification, security investigations, and audit reporting. |
Collaborate, reproduce, and manage knowledge
| Concept | What it is |
|---|---|
| A shared space for people to collaborate with access to the same set of code, data, results, and history. The primary way of organizing work in Domino. | |
| Domino keeps records of all the “inputs” for most deliverables and results you create in the platform: the code, data, and Compute Environment that produced a given result or Model or App. | |
| Domino indexes contents in the platform so you can find and reuse past work easily. | |
| A Project pre-configured with code, data, and other materials, to be used as a starting point for new Projects to help people use best practices and avoid re-inventing the wheel. | |
| Tags are shared, hierarchical categories for grouping and finding related work. Properties are typed fields that record facts about each asset. |
Architecture and administration
| Concept | What it is |
|---|---|
| Domino runs every workload as a container on Kubernetes. A cluster autoscaler adds compute capacity as demand grows. | |
| One Domino deployment can run workloads in several regions, clouds, or on-premises data centers, each a Data Plane attached to the same Control Plane. | |
| A versioned, shareable definition of the tools and packages a workload runs in, built from a Docker image, so everyone runs the same software. | |
| Administrator-defined compute profiles (CPU, memory, GPU) that you pick when launching a workload. Access can be limited to an organization. |