DataRobot Agent Workforce Platform overview¶

The DataRobot Agent Workforce Platform seamlessly orchestrates the full lifecycle of AI agents across a wide range of deployment environments. It is available on managed SaaS, virtual private cloud, or self-managed infrastructure—including on-prem and private GPU clouds, sovereign GPU clouds, air-gapped environments, and major public cloud platforms. It is co-engineered with NVIDIA and validated across infrastructure from Dell and Nebius. DataRobot enables faster, more secure deployment of AI solutions that drive real business impact.

Platform concepts¶

The following describes how the Agent Workforce Platform is built, including its architecture, governance layers, and deployment flexibility. In other words, it provides the principles behind where and how it runs.

Where the platform fits¶

Enterprise AI agents generally fall into three tiers:

1. Copilots—Assistants layered onto existing productivity tools.
2. Line-of-business agents—Agents embedded inside a single application.
3. Agent workforce agents—Agents that operate across an organization's data, systems, and processes, independent of any one cloud or application.

The Agent Workforce Platform is purpose-built for the third tier, with particular depth in environments where data residency, air-gapping, or infrastructure ownership rule out a single-cloud approach.

Platform architecture¶

The platform is organized into three layers:

• The Build layer is where agents are developed, evaluated, and prepared for deployment.
• The Operate (or Run) layer is where agents run, are monitored, and are orchestrated in production.
• The Govern layer spans both, applying AI, IT, and infrastructure governance from the first agent.

ガバナンス¶

Governance in DataRobot operates at three layers:

1. AI and agentic governance—Approval workflows and versioning before an agent ships, plus real-time moderation that catches bias, hallucinations, prompt injection, and PII leakage in production. Production monitoring also tracks drift, correctness, and tool-calling accuracy.

2. IT governance—Agent-specific identity, permissioning, and entitlements, with consistent lineage and state auditing across agents, tools, and applications. Agent identity integrates with existing enterprise identity infrastructure rather than replacing it.

3. Infrastructure governance—Gateways, fair-use policies, and cost controls that track GPU cost and utilization and keep spend predictable at scale.

The same governance, monitoring, and control apply wherever the platform runs.

Deployment flexibility¶

A core design principle of the Agent Workforce Platform is avoiding lock-in at the model, cloud, or orchestration layer. The same build, operate, and govern capabilities work consistently across managed SaaS, private cloud, self-managed, and fully air-gapped environments, so agents can be built once and deployed into whichever environment an organization's data and compliance requirements call for.

Inference and infrastructure¶

Run self-hosted inference on GPUs you already own, and choose the model for each task. The LLM Gateway gives you access to 70+ generative AI and embedding models without managing API keys, and NVIDIA NIM and NeMo microservices are built into the platform.

Delivery model¶

Production deployments are typically built with DataRobot's Center of Excellence—forward-deployed engineers who work alongside a customer's own team to build the first mission-critical agents, rather than relying solely on self-serve tooling. This is by design for the platform's target use cases: high-complexity, high-value agents in regulated or infrastructure-constrained environments, where getting from pilot to production has historically stalled.

Platform tools¶

The following sections describe the places and tooling in the DataRobot platform where you build, run, and govern generative and agentic AI.

ワークベンチ¶

Building great AI solutions—generative AI, predictive models, and agentic workflows—requires a lot of experimentation. If you're in the UI, use Workbench to quickly experiment, easily compare across experiments, and organize all your experiment assets in an intuitive Use Case container. ユーザーは、特定の問題を解決するために必要なすべてのものを1つの場所でグループ化、共有できます。 ワークベンチはユースケース別に構成され、各ユースケースには0個以上のデータセット、ベクターデータベース、プレイグラウンド、モデル、ノートブック、およびアプリケーションが含まれます。

レジストリ¶

Use Registry to create deployment-ready model and agent packages and generate compliance documentation for enterprise governance.

レジストリにより、すべてのAIアセットがドキュメント化され、バージョン管理下に置かれます。 With test results and metadata stored alongside each AI asset, you can deploy to production with confidence, regardless of whether the asset originated with DataRobot, was user-defined, was brought in from an external framework, or is an agent built with a third-party tool such as CrewAI, LangGraph, or LlamaIndex.

コンソール¶

Use Console to view the operating status of every deployed agent and model.

As your organization becomes more AI-driven, you'll have tens or even hundreds of task-specific models and agents running at once. Console provides a centralized hub for observing their performance, tracing agent decisions and tool calls back to their inputs, and configuring automated intervention and notification options to keep everything running smoothly and within policy. そのダッシュボードでは、デプロイされたモデルにアクセスして、さらに監視とリスク軽減を行うことができます。 It also provides access to prediction activities and allows you to view, create, edit, delete, or share serverless and external prediction environments.

ワンクリックでデプロイ¶

But where do deployments fit into the structure above? The deployment process is your model's or agent's seamless transition from Registry to Console. Once it's registered and tested in Registry, you'll have the option to deploy it with a single click.

Once you click to deploy, DataRobot's automation creates an API endpoint in your selected prediction or runtime environment and configures all the observability, monitoring, and governance controls for it.

これら4つのデプロイオプションがすべてサポートされています。

• DataRobotモデルをDataRobotサーバーレスまたは専用予測サーバーにデプロイ。
• DataRobotモデルを外部の予測サーバーにデプロイ
• カスタムモデルをDataRobotサーバーレスまたは専用予測サーバーにデプロイ。
• カスタムモデルを外部の予測サーバーにデプロイ

Agentic AI deployments follow the same registered, governed path into production.

Generative AI and agentic AI¶

DataRobot's generative AI capabilities build on its predictive AI experience, and its agentic AI capabilities build on both. Bring your favorite libraries and frameworks (CrewAI, LangGraph, LlamaIndex, or a generic template of your own), choose your LLMs and vector databases, and integrate third-party tools—then embed or deploy the result wherever it drives value for your business, with governance built in at every layer.

Inside DataRobot there is very little distinction between predictive, generative, and agentic AI. The same underlying platform (Workbench, Registry, Console) supports building, operating, and governing all three. But you'll find that tutorials and examples are still organized along these lines, since the workflows differ:

• エージェント型AI

• Generative AI (GenAI) includes text and image generation using foundation models, typically as a single LLM call or RAG pipeline.

• 予測AI

Agentic and Generative AI

Predictive AI

アプリケーション¶

DataRobot offers various approaches for building applications that allow you to share machine learning projects, generative AI workflows, and agents: custom applications, application templates, and no-code applications.

• カスタムアプリケーションは、カスタムコードを構築して実行するための簡単な方法です。
• カスタマイズ可能なアプリケーションテンプレートは、ユースケースをサポートするDataRobotのリソースをプログラムで生成することで、ユーザーを支援します。
• ノーコードアプリケーションでは、モデルを構築して評価することなく、DataRobotのコアサービスを利用できます。

データと会話するエージェントのアプリケーションテンプレートを試す

どちらのエクスペリエンスを選ぶべきか？¶

DataRobotで作業する場合、インターフェイスには2つの選択肢があります。

1. Log in to the platform via web browser and work with the graphical user interface (UI).
2. Access the platform programmatically with the REST API or Python client packages, or build agents with the DataRobot CLI in your own development environment (IDE, notebook, or coding agent of choice).

If you're leaning toward using DataRobot programmatically, explore the workflows in the UI first. DataRobotは、どちらのインターフェイスでも完全なアクセシビリティをお約束します。そのため、1つの選択肢に縛られることはありません。 以下を実行することが可能です。

• いつでもコードとUIを柔軟に切り替えることができます。
• 異なるオプションを使用している他のユーザーとシームレスにコラボレーションできます。
• When in code, use any development environment of your choice, and bring agents built elsewhere in for governance and deployment.

Try the UI¶

エクスペリメントベースの反復ワークフローを実現するワークベンチ、モデルの進化を追跡し、バージョン管理されたモデルを一元管理するためのレジストリ、デプロイされたモデルを監視・管理するコンソールで構成された、直感的なUIベースの製品です。 The UI provides a complete AI lifecycle platform, with broad interoperability and end-to-end capabilities for ML experimentation and production. It is also the gateway for creating GenAI experiments, Notebooks, and applications.

予測の基本ステップを試す

コード¶

Code offers a programmatic alternative for accessing DataRobot using the REST API or Python client packages, or the DataRobot CLI for agent development. You can also install DataRobot agentic skills in Cursor, Claude Code, Gemini CLI, and other coding agents to bring platform workflows into your IDE.

開発者クイックスタートを参照

ヘルプを得るには¶

以下のどこからでもヘルプを得ることができます。

次のステップ¶

Visit the How-tos for walkthroughs of a variety of project types:

• エージェント
• GenAI
• アプリケーション
• 時系列予測