The shift from prototyping to having brokers in manufacturing is the problem for AI groups as we glance towards 2026 and past. Constructing a cool prototype is simple: hook up an LLM, give it some instruments, see if it appears prefer it’s working. The manufacturing system, now that’s exhausting. Brittle integrations. Governance nightmares. Infrastructure wasn’t constructed for the complexities and nuances of brokers.
For AI builders, the problem has shifted from constructing an agent to orchestrating, governing, and scaling it in a manufacturing setting. DataRobot’s newest launch introduces a sturdy suite of instruments designed to streamline this lifecycle, providing granular management with out sacrificing velocity.
New capabilities accelerating AI agent manufacturing with DataRobot
New options in DataRobot 11.2 and 11.3 provide help to shut the hole with dozens of updates spanning observability, developer expertise, and infrastructure integrations.
Collectively, these updates concentrate on one objective: decreasing the friction between constructing AI brokers and operating them reliably in manufacturing.
Probably the most impactful areas of those updates embody:
- Standardized connectivity by way of MCP on DataRobot
- Safe agentic retrieval by way of Discuss to My Docs (TTMDocs)
- Streamlined agent construct and deploy by way of CLI tooling
- Immediate model management by way of Immediate Administration Studio
- Enterprise governance and observability by way of useful resource monitoring
- Multi-model entry by way of the expanded LLM Gateway
- Expanded ecosystem integrations for enterprise brokers
The sections that comply with concentrate on these capabilities intimately, beginning with standardized connectivity, which underpins each production-grade agent system.
MCP on DataRobot: standardizing agent connectivity
Brokers break when instruments change. Customized integrations develop into technical debt. The Mannequin Context Protocol (MCP) is rising as the usual to unravel this, and we’re making it production-ready.
We’ve added an MCP server template to the DataRobot group GitHub.
- What’s new: An MCP server template you’ll be able to clone, take a look at domestically, and deploy on to your DataRobot cluster. Your brokers get dependable entry to instruments, prompts, and sources with out reinventing the combination layer each time. Simply convert your predictive fashions as instruments which are discoverable by brokers.
- Why it issues: With our MCP template, we’re supplying you with the open commonplace with enterprise guardrails already inbuilt. Check in your laptop computer within the morning, deploy to manufacturing by afternoon.
Discuss to My Docs: Safe, agentic information retrieval
Everyone seems to be constructing RAG. Nearly no one is constructing RAG with RBAC, audit trails, and the power to swap fashions with out rewriting code.
The “Discuss to My Docs” software template brings pure language chat-style productiveness throughout all of your paperwork and is secured and ruled for the enterprise.
- What’s new: A safe, ruled chat interface that connects to Google Drive, Field, SharePoint, and native recordsdata. In contrast to fundamental RAG, it handles advanced codecs from tables, spreadsheets, multi-doc synthesis whereas sustaining enterprise-grade entry management.
- Why it issues: Your staff wants ChatGPT-style productiveness. Your safety staff wants proof that delicate paperwork keep restricted. This does each, out of the field.
Agentic software starter template and CLI: Streamlined construct and deployment
Getting an agent into manufacturing mustn’t require days of scaffolding, wiring companies collectively, or rebuilding containers for each small change. Setup friction slows experimentation and turns easy iterations into heavyweight engineering work.
To handle this, DataRobot is introducing an agentic software starter template and CLI, each designed to cut back setup overhead throughout each code-first and low-code workflows.
- What’s new: An agentic software starter template and CLI that allow builders configure agent elements by way of a single interactive command. Out-of-the-box elements embody an MCP server, a FastAPI backend, and a React frontend. For groups that choose a low-code strategy, integration with NVIDIA’s NeMo Agent Toolkit permits agent logic and instruments to be outlined solely by way of YAML. Runtime dependencies can now be added dynamically, eliminating the necessity to rebuild Docker photos throughout iteration.
- Why it issues: By minimizing setup and rebuild friction, groups can iterate sooner and transfer brokers into manufacturing extra reliably. Builders can concentrate on agent logic relatively than infrastructure, whereas platform groups preserve constant, production-ready deployment patterns.
Immediate administration studio: DevOps for prompts
As prompts transfer from experiments to manufacturing belongings, advert hoc enhancing rapidly turns into a legal responsibility. With out versioning and traceability, groups battle to breed outcomes or safely iterate.
To handle this, DataRobot introduces the Immediate Administration Studio, bringing software-style self-discipline to immediate engineering.
- What’s new: A centralized registry that treats prompts as version-controlled belongings. Groups can monitor modifications, evaluate implementations, and revert to steady variations as prompts transfer by way of improvement and deployment.
- Why it issues: By making use of DevOps practices to prompts, groups acquire reproducibility and management, making it simpler to transition from prototyping to manufacturing with out introducing hidden danger.
Multi-tenant governance and useful resource monitoring: Operational management at scale
As AI brokers scale throughout groups and workloads, visibility and management develop into non-negotiable. With out clear perception into useful resource utilization and enforceable limits, efficiency bottlenecks and price overruns rapidly comply with.
- What’s new: The improved Useful resource Monitoring tab supplies detailed visibility into CPU and reminiscence utilization, serving to groups determine bottlenecks and handle trade-offs between efficiency and price. In parallel, Multi-tenant AI Governance introduces token-based entry with configurable price limits to make sure honest useful resource consumption throughout customers and brokers.
- Why it issues: Builders acquire clear perception into how agent workloads behave in manufacturing, whereas platform groups can implement guardrails that forestall noisy neighbors and uncontrolled useful resource utilization as methods scale.
Expanded LLM Gateway: Multi-model entry with out credential sprawl
As groups experiment with agent conduct and reasoning, entry to a number of basis fashions turns into important. Managing separate credentials, price limits, and integrations throughout suppliers rapidly introduces operational overhead.
- What’s new: The expanded LLM Gateway provides assist for Cerebras and Collectively AI alongside Anthropic, offering entry to fashions equivalent to Gemma, Mistral, Qwen, and others by way of a single, ruled interface. All fashions are accessed utilizing DataRobot-managed credentials, eliminating the necessity to handle particular person API keys.
- Why it issues: Groups can consider and deploy brokers throughout a number of mannequin suppliers with out rising safety danger or operational complexity. Platform groups preserve centralized management, whereas builders acquire flexibility to decide on the correct mannequin for every workload.
New supporting ecosystem integrations
Jira and Confluence connectors: To energy your vector databases, DataRobot supplies a cohesive ecosystem for constructing enterprise-ready, knowledge-aware brokers.
NVIDIA NIM Integration: Deploy Llama 4, Nemotron, GPT-OSS, and 50+ GPU-optimized fashions with out the MLOps complexity. Pre-built containers, production-ready from day one.
Milvus Vector Database: Direct integration with the main open-source VDB, plus the power to pick distance metrics that really matter in your classification and clustering duties.
Azure Repos & Git Integration: Seamless model management for Codespaces improvement with Azure Repos or self-hosted Git suppliers. No guide authentication required. Your code stays centralized the place your staff already works.
Get hands-on with DataRobot’s Agentic AI
In the event you’re already a buyer, you’ll be able to spin up the GenAI Check Drive in seconds. No new account. No gross sales name. Simply 14 days of full entry inside your present SaaS setting to check these options together with your precise information.
Not a buyer but? Begin a 14-day free trial and discover the total platform.
For extra data, please go to our Model 11.2 and Model 11.3 launch notes within the DataRobot docs.