Back to vendors
D

DataRobot

Also known as: DataRobot Agent Workforce Platform, DataRobot Agentic AI

Visit site
Entry priceQuote only for production agents, 30 day trialFull pricing detail

DataRobot provides a platform for enterprise teams to build agents, deploy them, and manage them in production. Developers can start with an application template or use their own code and agent framework. The platform registers versions, serves applications, and records traces of model and tool calls. Administrators can define approval requirements before deployment and monitor applications after they go live. DataRobot supports managed hosting, private cloud deployments, and installation on customer operated infrastructure.

DataRobot gives enterprise teams one environment for developing and running agents alongside predictive machine learning and generative AI applications, sold as the Agent Workforce Platform. Developers can start from an agent template or bring code built with frameworks such as CrewAI, LangGraph, and LlamaIndex. Workbench groups experiments, data, and models by use case.

Registry stores the workflow versions that a team prepares for deployment, and Console manages workloads serving real traffic. A platform team can connect services such as Snowflake, SAP, or Salesforce through tools that the agent can call while it is working. A library of prebuilt tools covers data queries, time series forecasts, multimodal retrieval, video search and research reports, and DataRobot also sells foundational, business and purpose built agents.

Agents can retrieve documents from a vector database managed by DataRobot or connect to an existing service such as Pinecone or PostgreSQL. The memory service can keep conversational context across sessions, with limits on the retention of stored events. Teams can configure each application with the DataRobot model gateway, an external model provider, or self hosted inference, including NVIDIA NIM.

Developers can run a benchmark dataset against an endpoint, score it with LLM as a judge metrics and inspect which parts of the workflow account for an incorrect response. In production, operators can view traces, configure moderation and guard models, and route changes to a deployment through an administrator approval workflow.

The platform runs as managed software, in a virtual private cloud, or on infrastructure the buyer operates. Self managed deployments can use private GPU clusters, sovereign clouds and air gapped environments, and DataRobot is part of NVIDIA's Enterprise AI Factory reference design. Its cloud control environment holds SOC 2 Type II and ISO 27001. Buyers can try a limited hosted environment for 30 days.

Production agent workflows require premium enablement through a DataRobot representative. The platform also gives operators a way to monitor and govern applications built outside DataRobot when those applications send the required deployment information and traces. Gartner has named DataRobot a Leader in its Magic Quadrant for Data Science and Machine Learning Platforms three times.

Chevron builds agents for inspection work in hazardous environments on DataRobot. Norfolk Iron & Metal made its AI deployments 18 times faster for predictive workflows such as inventory forecasting, NetApp combines Snowflake, SQL and S3 data to speed up forecasting, and Etisalat UAE offers agentic AI applications to its own customers through a DataRobot partnership. The same platform can connect those forecasts to a tool used by an agent.

An operations team could use a demand estimate to prepare a replenishment recommendation, then inspect the trace to see which forecast, business records, and instructions informed the result. Teams with stricter infrastructure requirements can keep inference and application traffic within a private environment. DataRobot provides engineering support for initial production deployments, while customer developers can also package their own workflows and deploy them through the command line. For teams that operate several applications, the registry provides a common place to find the workflow version and its test results.

Vendor details

Canonical URL

https://www.datarobot.com

Category

Agent builder

Subcategory

Enterprise agent development, deployment, and governance

Funding status

DataRobot, Inc. is a privately held software company.

Company status

independent

Use cases & customers

Primary use cases

Build and deploy agent applications from templates or custom codeMonitor and govern deployed agents and AI modelsIntegrate retrieval and predictive models into agent workflows

Target customers

Enterprise AI and machine learning engineering teamsPlatform operations teams governing deployed agentsOrganizations deploying agents in private or regulated environments

Deployment options

Managed SaaSVirtual private cloudSelf managed infrastructure including private GPU and air gapped environments

Integrations

Supported development frameworks include CrewAI, LangGraph, and LlamaIndex. Agents can use external business systems, connected vector databases, and separately deployed tools.

In practice

A supply chain team exposes a DataRobot demand forecast as a tool and builds an agent that combines the forecast with current warehouse records.

A platform engineer deploys a LangGraph application into a private cloud, reviews its benchmark results, and sends the deployment for administrator approval.

An operations team inspects traces for a document agent, identifies a retrieval step that selected the wrong source, and deploys a corrected workflow version.

Agentic Index coverage score

12.5 / 14 capabilities · 89%

Integrations & Tool Calling Full

Developers deploy tools separately from the workflow, give them the access credentials they need and call them from agents, so an agent can work in outside business systems and services while it runs. Each deployed tool is monitored and moderated in Console as its own managed component.

Sourcedocs.datarobot.com/en/docs/agentic-ai/agentic-develop/agentic-tools.htmlread 2026-11-20

Workflow Orchestration Full

The platform ships starter workflows for several agent frameworks and brings their applications into a managed development and production lifecycle. Teams package agent logic in code, register a workflow version, then run it as a deployed application. The platform supports coordinated agent components and handles their deployment even when the framework defines the workflow in code.

Sourcedocs.datarobot.com/latest/en/docs/agentic-ai/index.htmlread 2026-11-20

Knowledge Grounding & RAG Full

The platform builds and versions vector databases from files and data attached to a use case, then gives developers a way to use the selected version in a deployed retrieval pipeline. A customer can choose DataRobot hosted FAISS, a brought deployment, or a directly connected external store. Chunking and embedding are supported in the hosted workflow. Developers can reuse a registered vector database in another use case.

Sourcedocs.datarobot.com/en/docs/agentic-ai/vector-database/vector-dbs.htmlread 2026-11-20

Human Oversight & Guardrails Full

Deployment changes and new agent versions can require human approval through approval workflow APIs and administrator controls, so a change stays a draft until someone approves it. Configurable runtime moderation and guard models add checks on inputs and requested output once an agent is live.

Sourcedocs.datarobot.com/latest/en/docs/get-started/day0/orientation.htmlread 2026-11-20

Security, Identity & Governance Full

DataRobot's cloud services hold SOC 2 Type II and ISO 27001. Administrators can use single sign on and role based controls, including custom roles, and agents carry identities and entitlements of their own. Deployment controls let an enterprise restrict where application code and data run.

Sourcedatarobot.com/trustcenterread 2026-11-20

Observability & Auditability Full

The experimentation dashboard records traces with OpenTelemetry and lets a developer inspect the sequence of model operations and tool calls. Production operators inspect agent traces and monitoring records in Console under the associated deployment. Traces show how steps relate, and an operator can open any step when an output or tool action needs investigation, with lineage back to inputs, tools and decisions.

Sourcedocs.datarobot.com/latest/en/docs/agentic-ai/cli/local-experimentation/experimentation-trace-analysis.htmlread 2026-11-20

Memory & State Persistence Full

A memory service stores conversation events for agent applications in addressable memory spaces and sessions, so later session requests can retrieve them. Administrators set how long events are retained. Self managed installs back the service with PostgreSQL and pgvector, kept apart from the vector database that grounds answers.

Sourcedocs.datarobot.com/latest/en/docs/install/installation/advanced-configuration/memory-service.htmlread 2026-11-20

Deployment & Data Residency Full

Build, deployment and oversight run on the managed platform, in a virtual private cloud or on customer managed infrastructure. Self managed installs run on premises on the customer's GPU hardware, in air gapped networks and in sovereign or private clouds. The infrastructure choice changes how the buyer provisions resources, while the application lifecycle stays in the platform.

Sourcedocs.datarobot.com/latest/en/docs/get-started/day0/orientation.htmlread 2026-11-20

Prebuilt Agents / Templates / Packs Full

Deployable starter applications include an agentic application with a FastAPI backend and React interface, a data assistant and a multi agent document assistant, each able to run locally and deploy to DataRobot by command. Foundational, business and purpose built agents are sold beside them, with some SAP packages limited to premium access.

Sourcedocs.datarobot.com/latest/en/docs/reference/gen-ai-ref/agentic-repos.htmlread 2026-11-20

Triggers & Channel Coverage Partial

Registered agent workflows run behind a serving endpoint that another application or process calls, and a single API call can start predictions and batch jobs on demand. Managed Airflow keeps models and tools current through automated pipelines, while agents have no scheduler or event subscription of their own to wake them.

Sourcedocs.datarobot.com/en/docs/agentic-ai/agentic-deploy/agentic-workflow-deploy.htmlread 2026-11-20

Model Flexibility & Routing Full

Agent developers can use the default DataRobot model gateway or configure another provider through LiteLLM connectors for Azure OpenAI, AWS Bedrock, Google, Cohere, Hugging Face and Ollama. Customers with the required infrastructure can also serve models through NVIDIA NIM or other self hosted inference, and the experimentation tools compare models side by side.

Sourcedocs.datarobot.com/latest/en/docs/agentic-ai/agentic-develop/agentic-llm-providers.htmlread 2026-11-20

APIs / SDKs / MCP Extensibility Full

A Python client and a REST API expose core platform operations, beside an open source command line for agent development and deployment. Developers can automate management workflows and call deployed endpoints from a separate application. The same interfaces authenticate a caller, address a deployment and run agents packaged in the customer's own project.

Sourcedocs.datarobot.com/latest/en/docs/api/dev-learning/api-quickstart.htmlread 2026-11-20

Testing, Debugging & Optimization Full

Batch evaluation runs a defined dataset against a selected agent endpoint through a chosen scoring pipeline, including LLM as a judge metrics. The experimentation interface shows each run, and CLI options let developers validate inputs without running the judge. Teams get a repeatable evaluation before changing a production deployment, separate from watching live traces.

Sourcedocs.datarobot.com/latest/en/docs/agentic-ai/cli/local-experimentation/experimentation-evaluation.htmlread 2026-11-20

Browser / Computer-use Not documented

Agents run as code beside separately deployed tools, and those tools can call third party services or customer systems. A developer edits code in a browser workspace, but the platform runs no managed browser or desktop session for an agent to drive. The tools catalog holds agent functions and deployed services.

Sourcedocs.datarobot.com/en/docs/agentic-ai/agentic-develop/agentic-tools.htmlread 2026-11-20

The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded

Pricing

Quote only for production agents, 30 day trial

Custom enterprise agreement for premium agent capabilities

Trial available

What is public

DataRobot offers a free 30 day hosted trial. The trial includes selected agent development and workflow capabilities.

Billing mechanics

Production agent features require premium enablement. A DataRobot representative handles contracts and provisioning.

Variable cost rationale

Production agent pricing is set in each enterprise agreement, so how spend moves with agent traffic, model calls and GPU use depends on contract terms that are not public.

Additional watchouts

The trial has limits on model calls and excludes some deployment and compute options.

Sales call required

Yes, required for paid access

Free / trial

30 day hosted trial with limited agent functionality

Missing data

Subscription rates, model use charges and private deployment costs come only through a DataRobot representative.

Agentic Index verified 2026-11-20

Alternatives to DataRobot

Other products that do the work DataRobot does, and what each one is.

Contact us

Found a vendor we missed? Have feedback on the index? We'd love to hear from you.