DataRobot and Nebius have partnered to introduce AI Manufacturing facility for Enterprises, a joint answer designed to speed up the event, operation, and governance of AI brokers. This platform permits brokers to achieve manufacturing in days, moderately than months.
AI Manufacturing facility for Enterprises offers a scalable, cost-effective, ruled, and managed enterprise-grade platform for brokers. It achieves this by combining DataRobot’s Agent Workforce Platform: essentially the most complete, versatile, safe, and enterprise-ready agent lifecycle administration platform, with Nebius’ purpose-built cloud infrastructure for AI.
Our partnership
Nebius: The aim-built cloud for AI
The problem at present is that general-purpose cloud platforms typically introduce unpredictable efficiency, latency, and a “virtualization tax” that cripples steady, production-scale AI.
To unravel this, DataRobot is leveraging Nebius AI Cloud, a GPU cloud platform engineered from the {hardware} layer up particularly to ship the bare-metal efficiency, low latency, and predictable throughput important for sustained AI coaching and inference. This eliminates the “noisy-neighbor” downside and ensures your most demanding agent workloads run reliably, delivering predictable outcomes and clear prices.
Nebius’ Token Manufacturing facility augments the providing by offering a pay-per-token mannequin entry layer for key open-source fashions, which prospects can use throughout agent constructing and experimentation, after which deploy the identical fashions with DataRobot when operating the brokers in manufacturing.
DataRobot: Seamlessly construct, function, and govern brokers at scale
DataRobot’s Agent Workforce Platform is essentially the most complete Agent Lifecycle Administration platform that permits prospects to construct, function, and govern their brokers seamlessly.
The platform affords two major parts:
- An enterprise-grade, scalable, dependable, and cost-effective runtime for fashions and brokers, that includes out-of-the-box governance and monitoring.
- A straightforward-to-use agent builder setting that permits prospects to seamlessly construct production-ready brokers in hours, moderately than days or months.
Complete enterprise-grade runtime capabilities
- Scalable, cost-effective runtime: Options single-click deployment of fifty+ NIMs and Hugging Face fashions with autoscaling or deploy any containerized artifacts through Workload API (each with inbuilt monitoring/governance), optimized utilization via endpoint stage multi-tenancy (token quota), and high-availability inferencing. You may deploy containerized brokers, purposes or different composite techniques constructed utilizing a mixture of say LLMs, area particular libraries like PhysicsNemo, cuOpt and so forth., or your individual proprietary fashions, with a single command utilizing Workload API.
- Governance and monitoring: Supplies the {industry}’s most complete out-of-the-box metrics (behavioral and operational), tracing capabilities for agent execution paths, full lineage/versioning with audit logging, and industry-leading governance towards Safety, Operational, and Compliance Dangers with real-time intervention and automatic reporting.
- Safety and identification: Consists of Unified Identification and Entry Administration with OAuth 2.0, granular RBAC for least-privilege entry throughout assets, and safe secret administration with an encrypted vault.
Complete enterprise-grade agent constructing capabilities
- Builder instruments: Help for common frameworks (Langchain, Crew AI, Llamaindex, Nvidia NeMo Agent Toolkit) and out-of-the-box assist for MCP, authentication, managed RAG, and information connectors. Nebius token manufacturing unit integration permits on-demand mannequin use throughout the construct.
- Analysis & tracing: Business-leading analysis with LLM as a Choose, Human-in-the-Loop, Playground/API, and agent tracing. Provides complete behavioral (e.g., process adherence) and operational (latency, value) metrics, plus customized metric assist.
- Out-of-the field manufacturing readiness: Enterprise hooks summary away infrastructure, safety, authentication, and information complexity. Brokers deploy with a single command; DataRobot handles element deployment with embedded monitoring and governance at each the complete agent and particular person element/instrument ranges.
Construct and deploy utilizing the AI Manufacturing facility for Enterprises
Wish to take brokers you will have constructed elsewhere, and even open supply {industry} particular fashions and deploy them in a scalable, safe and ruled method utilizing the AI Manufacturing facility? Or would you prefer to construct brokers with out worrying in regards to the heavy lifting of constructing them manufacturing prepared? This part will present you find out how to do each.
1. DataRobot STS on Nebius
DataRobot Single-Tenant SaaS (STS) is deployed on Nebius Managed Kubernetes and might be backed by GPU-enabled node teams, high-performance networking, and storage choices acceptable for AI workloads.For DataRobot deployments, Nebius is a high-performance low value setting for agent workloads. Devoted NVIDIA clusters (H100, H200, B200, B300, GB200 NVL72, GB300 NVL72) allow environment friendly tensor parallelism and KV-cache-heavy serving patterns, whereas InfiniBand RDMA helps high-throughput cross-node scaling. The DataRobot/Nebius partnership offers a strong AI infrastructure:
- Managed kubernetes with GPU-aware scheduling simplifies STS set up and upgrades, pre-configured with NVIDIA operators.
- Devoted GPU employee swimming pools (H100, B200, and so forth.) isolate demanding STS companies (LLM inference, vector databases) from generic CPU-only workloads.
- Excessive-throughput networking and storage assist giant mannequin artifacts, embeddings, and telemetry for steady analysis and logging.
- Safety and tenancy is maintained: STS makes use of devoted tenant boundaries, whereas Nebius IAM and community insurance policies meet enterprise necessities.
- Constructed-in node well being monitoring proactively identifies and addresses GPU/community points for secure clusters and smarter upkeep.
2. Ruled, monitored mannequin inference deployment
The problem with GenAI isn’t getting a mannequin operating; it’s getting it operating with the identical monitoring, governance, and safety your group expects. DataRobot’s NVIDIA NIM integration deploys NIM containers from NGC onto Nebius GPUs in 4 clicks:
- In Registry > Fashions, click on Import from NVIDIA NGC and browse the NIM gallery.
- Choose the mannequin, evaluation the NGC mannequin card, and select a efficiency profile.
- Assessment the GPU useful resource bundle robotically beneficial primarily based on the NIM’s necessities.
- Click on Deploy, choose the Serverless setting, and deploy the mannequin.
Out-of-the-box observability and governance for deployed fashions
- Automated monitoring & threat evaluation: Leverage the NeMo Evaluator integration for mannequin faithfulness, groundness, and relevance scoring. Mechanically scan for Bias, PII, and Immediate Injection dangers.
- Actual-time moderation & deep observability: DataRobot affords a platform for NIM moderation and monitoring. Deploy out-of-the-box guards for dangers like PII, Immediate Injection, Toxicity, and Content material Security. OTel-compliant monitoring offers visibility into NIM operational well being, high quality, security, and useful resource use.
- Enterprise governance & compliance: DataRobot offers the executive layer for protected, organization-wide scaling. It robotically compiles monitoring and analysis information into compliance documentation, mapping efficiency to regulatory requirements for audits and reporting.
3. Agent deployment utilizing the Workload API
An MCP instrument server, a LangGraph agent, a FastAPI backend, composite techniques constructed utilizing mixture of say LLMs and area particular libraries like cuOpt, PhysicsNemo and so forth; these are containers, not fashions, and so they want their very own path to manufacturing. The Workload API provides you a ruled endpoint with autoscaling, monitoring, and RBAC in a single API name.
curl -X POST "${DATAROBOT_API_ENDPOINT}/workloads/"
-H "Authorization: Bearer ${DATAROBOT_API_TOKEN}"
-H "Content material-Kind: utility/json"
-d '{
"identify": "agent-service",
"significance": "HIGH",
"artifact": {
"identify": "agent-service-v1",
"standing": "locked",
"spec": {
"containerGroups": [{
"containers": [{
"imageUri": "your-registry/agent-service:latest",
"port": 8080,
"primary": true,
"entrypoint": ["python", "server.py"],
"resourceRequest": {"cpu": 1, "reminiscence": 536870912},
"environmentVars": [
],
"readinessProbe": {"path": "/readyz", "port": 8080}
}]
}]
}
},
"runtime": {
"replicaCount": 2,
"autoscaling": {
"enabled": true,
"insurance policies": [{
"scalingMetric": "inferenceQueueDepth",
"target": 70,
"minCount": 1,
"maxCount": 5
}]
}
}
}'
The agent is straight away accessible at /endpoints/workloads/{id}/ with monitoring, RBAC, audit trails, and autoscaling.
Out-of-the-box observability and governance for deployed agentic workloads
DataRobot drives the AI Manufacturing facility by offering strong governance and observability for agentic workloads:
- Observability (OTel Commonplace): DataRobot standardizes on OpenTelemetry (OTel): logs, metrics, and traces—to make sure constant, high-fidelity telemetry for all deployed entities. This telemetry seamlessly integrates with current enterprise observability stacks, permitting customers to watch essential dimensions, together with:
- Agent-specific metrics: Comparable to Agent Job Adherence and Agent Job Accuracy.
- Operational well being and useful resource utilization.
- Tracing and Logging: OTel-compliant tracing interweaves container-level logs with execution spans to simplify root trigger evaluation inside advanced logic loops.
- Governance and Entry Management: DataRobot enforces enterprise-wide authentication and authorization protocols throughout deployed brokers utilizing OAuth-based entry management mixed with Function-Based mostly Entry Management (RBAC).
4. Enterprise-ready agent constructing capabilities
A complete toolkit for each builder with the DataRobot Agent Workforce Platform on Nebius
The DataRobot Agent Workforce Platform helps builders construct brokers quicker by extending current flows. Our builder kits assist advanced multi-agent workflows and single-purpose bots, accommodating varied instruments and environments.
Our equipment contains native assist contains:
- Open supply frameworks: Native integration with LangChain, CrewAI, and LlamaIndex.
- NAT (Node Structure Tooling): DataRobot’s framework for modular, node-based agent design.
- Superior requirements: Expertise, MCP (Mannequin Context Protocol) for information/instrument interplay, and strong Immediate Administration for versioning/optimization.
The Nebius benefit: DataRobot’s Agent Workforce Platform integrates with the Nebius Token Manufacturing facility, permitting builders to eat fashions like Nemotron 3 (and any open supply mannequin) on a pay-per-token foundation throughout the experimental part. This permits speedy, low-cost iteration with out heavy infrastructure provisioning. As soon as perfected, brokers can seamlessly transition from the Token Manufacturing facility to a devoted deployment (e.g., NVIDIA NIM) for enterprise scale and low latency.
Getting Began: Constructing is straightforward utilizing our Node Structure Tooling (NAT). You outline agent nodes as structured, testable steps in YAML.
First, join your deployed LLM within the Nebius token elements to DataRobot
Add DataRobot deployment to you agentic starter utility within the DataRobot CLI
features:
planner:
_type: chat_completion
llm_name: datarobot_llm
system_prompt: |
You're a content material planner. You create transient, structured outlines for weblog articles.
You determine an important factors and cite related sources. Maintain it easy and to the purpose -
that is simply a top level view for the author.
Create a easy define with:
1. 10-15 key factors or details (bullet factors solely, no paragraphs)
2. 2-3 related sources or references
3. A quick urged construction (intro, 2-3 sections, conclusion)
Do NOT write paragraphs or detailed explanations. Simply present a targeted record.
author:
_type: chat_completion
llm_name: datarobot_llm
system_prompt: |
You're a content material author working with a planner colleague.
You write opinion items primarily based on the planner's define and context. You present goal and
neutral insights backed by the planner's data. You acknowledge when your statements are
opinions versus goal details.
1. Use the content material plan to craft a compelling weblog put up.
2. Construction with an enticing introduction, insightful physique, and summarizing conclusion.
3. Sections/Subtitles are correctly named in an enticing method.
4. CRITICAL: Maintain the whole output beneath 500 phrases. Every part ought to have 1-2 transient paragraphs.
Write in markdown format, prepared for publication.
content_writer_pipeline:
_type: sequential_executor
tool_list: [planner, writer]
description: A instrument that plans and writes content material on the requested subject.
function_groups:
mcp_tools:
_type: datarobot_mcp_client
authentication:
datarobot_mcp_auth:
_type: datarobot_mcp_auth
llms:
datarobot_llm:
_type: datarobot-llm-component
workflow:
_type: tool_calling_agent
llm_name: datarobot_llm
tool_names:
- content_writer_pipeline
- mcp_tools
return_direct:
- content_writer_pipeline
system_prompt:
Select and name a instrument to reply the question.
Analysis capabilities: The “how-to”
Constructing is barely half the battle; realizing if it really works is the opposite. Our analysis framework strikes past easy “thumbs up/down” and into data-driven validation.
To judge your agent, you’ll be able to:
- Outline a check suite: Add a “golden dataset” of anticipated queries and ground-truth solutions.
- Automated metrics: Run your agent towards built-in evaluators for faithfulness, relevance, and toxicity.
- LLM-as-a-Choose: Use a “critic” mannequin to attain agent responses primarily based on customized rubrics (e.g., “Did the agent comply with the model’s tone of voice?”).
- Aspect-by-side comparability: Run two variations of your agent (e.g., one utilizing NAT and one utilizing LangChain) towards the identical dataset to match value, latency, and accuracy in a single dashboard.
Enterprise hooks: Deployment-ready from day one
We automate the “enterprise tax” (safety, logging, auth) that separates notebooks from manufacturing companies by embedding construct “hooks”:
- Observability: Computerized OTel-compliant tracing captures each step with out boilerplate.
- Identification & auth: Constructed-in OAuth 2.0 and Service Accounts guarantee brokers use the person’s precise permissions when calling inside APIs (CRM, ERP), sustaining strict safety.
- Manufacturing hand-off: Deployment packages the setting, parts, and auth hooks right into a safe, ruled container, making certain a constant agent from dev to manufacturing. Complicated brokers are autoparsed into orchestrated containers for granular monitoring whereas deployed as a single pipeline entity.
Ruled, scalable inference
The DataRobot and Nebius partnership delivers a validated, enterprise-ready deployment stack for agentic AI constructed on NVIDIA accelerated computing. For groups transferring past experimentation, it offers a ruled and scalable path to sustained manufacturing inference.
Nebius and DataRobot shall be showcasing this answer at NVIDIA GTC 2026, happening March 16-19 in San Jose, California.
Learn the chief abstract weblog
Join with DataRobot (sales space #104) and Nebius (sales space #713) at GTC 2026
