Glossary

Key terms and concepts used throughout the App Framework documentation.

A

Answer file: The YAML file written to .datarobot/answers/COMPONENT-NAME.yml when you apply a component. It records the choices you made during the copier wizard — template name, code name, port, LLM type, and so on. Passing this file to dr component update or uvx copier update -a replays your configuration non-interactively, enabling automated updates without re-answering the wizard.
App Framework: The DataRobot Application Framework for accelerating and making production grade agentic AI applications.
af-component-* (naming convention): The naming convention for all App Framework component repositories hosted in the datarobot-community GitHub organization. Each repository named af-component-* contains a single component — for example, af-component-base, af-component-fastapi-backend, af-component-agent. Search for af-component in the datarobot-community org to browse the full catalog.

C

Component: A copier-style template — a top-level folder or file set — that you layer into an application and is built on af-component-base to add one capability. Each component lives in its own repository named af-component-*, runs an interactive wizard when applied, defines its own dependencies in copier-module.yaml and records the answers to an answer file in .datarobot/answers/. Components are the primary building block of every App Framework application.
copier-module.yaml: An App Framework specific configuration file that defines the upstream and possible peer dependencies of an af-component.
copier-watch: A development utility (tools/copier-watch/copier-watch.py) that auto-applies component template changes to a destination recipe as you edit source files. On each save, it amends a local commit on the source repository and runs copier update on the destination — then resets and re-applies the full changeset from scratch, so you can iterate on Jinja templates without pushing remote commits. See the Developer Guide for usage and typical workflow.
Copier: The open-source templating tool (copier.readthedocs.io) that the App Framework uses to apply and update component templates. Unlike Yeoman or Cookiecutter, Copier has Git-native update semantics: it records the source template's commit hash in the answer file and can merge upstream template changes into your recipe with copier update, reducing merge conflicts on re-application.
Core library package: An optional shared core/ Python package generated by af-component-base when you answer yes to the "Include core library" wizard question. The package lives at the recipe root and provides common utilities — auth helpers, config loading, shared types — that other components in the same multicomponent recipe can import rather than duplicating code. Enable it when your recipe will include more than one backend component that needs shared logic.
Custom Application: A DataRobot platform resource that hosts and serves a web application — frontend and backend — within the DataRobot environment. App Framework FastAPI and React components deploy to DataRobot as Custom Applications via the Pulumi infrastructure declared in infra/. See the Custom Applications documentation for platform details.

D

DataRobot CLI (dr): The official DataRobot command-line interface — a standalone binary for macOS, Linux, and Windows that wraps authentication, environment setup, component workflows, and deployment into a single unified tool. It is the recommended entry point for all App Framework operations. See cli.datarobot.com for full documentation.
datarobot-pulumi-utils: A DataRobot Python library that extends the pulumi-datarobot provider with CustomResources, helper utilities, and App Framework–specific abstractions for declaring platform resources in Pulumi programs. Source: github.com/datarobot-oss/datarobot-pulumi-utils. See also the Pulumi DataRobot provider documentation.
.datarobot/ directory: The hidden configuration directory created by af-component-base at the root of every recipe. It stores DataRobot CLI configuration, the .env template, and all answer files under .datarobot/answers/. The CLI reads this directory for authentication, environment setup, and component update operations.
Declarative API: DataRobot's infrastructure-as-code offering — Terraform and Pulumi providers that let you define DataRobot platform resources (Use Cases, LLM deployments, vector databases, applications) as code. The App Framework uses the Pulumi Python provider and datarobot-pulumi-utils to manage all infrastructure. See the Declarative API documentation.

F

Foundation application template: One of DataRobot's official open-source full-stack application templates built using App Framework components. These are production-grade reference implementations — including Talk to My Docs, Talk to My Data, and the DataRobot Agent Application — that developers can fork and customize. See github.com/datarobot-community for the full catalog.

I

infra/ directory: Part of the af-component-base scaffold that contains all Pulumi infrastructure-as-code for a recipe. Each component contributes its own infra/infra/COMPONENT.py file declaring the DataRobot platform resources it needs — vector databases, LLM deployments, models, agents, and custom applications. The base Pulumi.yaml ties these files together and is the entry point for dr task deploy.

L

LLM Deployment: A DataRobot platform resource that hosts a language model and exposes it as an API endpoint within the DataRobot environment. The af-component-llm component provisions an LLM Deployment backed by an LLM Blueprint and wires it to the LLM Gateway. See the Deploy an LLM documentation for platform details.
LLM Gateway: DataRobot's managed proxy for routing requests to language models — including OpenAI, Anthropic, Azure OpenAI, and others — with built-in guardrails, logging, and governance. The af-component-llm component configures the LLM Gateway as the default integration. See the LLM Gateway documentation.
Low-code YAML agents vs. code-based agents: Two modes offered by af-component-agent. The NeMo Agent Toolkit (NAT) variant (nat) lets you define agent behavior, tools, and conversation flows in declarative YAML — no Python required (NeMo Agent Toolkit). This is the "low-code" path and is well-suited for straightforward Q&A or task-routing agents. The code-based variants — CrewAI (crewai), LangGraph (langgraph), and LlamaIndex (llamaindex) — give you full Python control over agent logic, state, memory, and tool use, at the cost of more boilerplate. Choose NAT when you want to move fast and avoid framework complexity; choose a code-based framework when you need fine-grained orchestration or custom tooling.

N

NeMo Agent Toolkit (NAT): NVIDIA's open-source framework for building low-code, YAML-defined agent pipelines. Agents, tools, and conversation flows are declared in structured YAML configuration files rather than Python code, making it accessible to developers who prefer a configuration-driven approach. In the App Framework, NAT is available as the nat framework option in af-component-agent. See the NeMo Agent Toolkit documentation.

P

pulumi-datarobot: The official DataRobot Pulumi provider — a Pulumi SDK that lets you manage DataRobot platform resources (use cases, deployments, playgrounds, custom applications, vector databases) as infrastructure-as-code in Python, TypeScript, Go, and .NET. Source: github.com/datarobot-community/pulumi-datarobot. Registry: pulumi.com/registry/packages/datarobot.

R

Recipe: A Git repository that is a composition of one or more af-component-* components combined into a working application. Recipes are designed to be forked — either from a foundation application template or built from scratch by applying components with the CLI. All application code, infrastructure, and answer files live together in one repository.
Repeatable component: An App Framework component that can be applied more than once to the same recipe under different names. For example, af-component-agent can be applied twice to deploy two independent agents within the same application, and af-component-fastapi-backend can be applied multiple times for distinct microservices. Each application produces a separate answer file.

T

Taskfile.yaml: A task runner configuration file — similar to a Makefile or mise task file — defined using the Task runner (go-task). Generated by af-component-base, it provides tab-complete–friendly commands with automatic .env loading and cross-component task composition via dr task compose. See taskfile.dev for Task runner documentation.

U

Use case: A DataRobot platform resource that acts as the organizational container for all resources belonging to a project — deployments, playgrounds, vector databases, registered models, and custom applications. The af-component-base Pulumi configuration creates a Use Case as the root resource, and all other components attach their resources to it. See the Use Cases documentation.