AI agents are moving from hype to necessity. No longer just clever assistants, they’re evolving into systems that act on data, automate decisions, and power cross-functional workflows. But as AI agents proliferate across the organization, a familiar challenge is emerging: fragmentation.
What begins as a promising use case quickly becomes a patchwork of disconnected tools, homegrown scripts, and LLM integrations that no one can fully monitor, let alone govern. Teams experiment in silos. IT loses visibility. And, before long, that initial momentum gets buried under technical debt and operational risk.
That’s not innovation. That’s agent sprawl.
To scale AI agents across the enterprise, you need more than a tool to build agents. You need a platform that makes agents a durable part of how your business works: grounded in your data, embedded in your workflows, and governed by design.
That's what Dataiku — The Universal AI Platform™ — does best. Over the past year, we've helped customers scale beyond 1,000 GenAI use cases, supporting their evolution from AI to GenAI, enabling them to stay at the forefront of innovation.
Now, we’re taking the next step — helping organizations bring AI agents into the heart of their operations with the same control, flexibility, and trust they’ve come to expect from Dataiku.
Introducing AI Agents With Dataiku
Today, we’re excited to unveil the next evolution of The Universal AI Platform™: the ability to create, connect, and control AI agents — at scale, and with confidence.
Unlike open source toolkits, embedded assistants limited to specific apps, or vendor-bound cloud frameworks, AI agents with Dataiku deliver centralized governance, agnostic deployment, native data integration, and continuous optimization from day one.
Here’s what sets them apart:
- Central creation for every user – True to Dataiku tradition, business analysts get a no‑code canvas, developers get full Python and LangChain support. Both environments share the same enterprise layer for security, lineage, and audit, ensuring that governance starts on day one.
- Multi‑model orchestration – Power agents with models from any source, including OpenAI, Anthropic, Mistral, or self-hosted — and switch providers without rewriting code, so you’re never locked to a single LLM vendor.
- Multi‑agent coordination – The Agent Connect hub lets IT deploy and route all conversational agents from a single console, preventing “agent sprawl” across teams.
- Built‑in evaluation & observability – Reuse Dataiku pipelines to score prompts, trace every tool call, and trigger alerts on drift or anomalous behavior.
- End‑to‑end governance – Agents are governed alongside your other models in Dataiku, enabling sign‑offs, registries, and risk scoring to ensure nothing reaches production without a formal review.
Build AI Agents That Work for Your Business
Dataiku supports two flexible paths for creating agents, all within the same platform. Business users can build agents using a no-code visual interface and natural language prompts, while developers and data scientists have full flexibility with LangChain, Python, and custom APIs. Both approaches can be tailored to your business logic, systems, and data.
Regardless of how they’re built, all agents share the same infrastructure for connections, control, and monitoring. That means no duplication, no silos — just faster iteration and easier scaling.
To further accelerate development, Dataiku provides pre-built and customizable agent tools. These are modular components that perform common actions, such as web search, data lookups, and sending emails. These tools can be extended to work with existing agents from platforms like Salesforce, ServiceNow, or Microsoft Copilot, allowing Dataiku agents to act as orchestrators or collaborators within broader enterprise ecosystems.
Agent tools can be used as-is, customized with prompting, or developed in code — and once created, they can be reused across agents and teams. It's a way to build smarter and faster, while increasing consistency across agent behavior.
Orchestrate Multi-Agent Systems With Agent Connect
As organizations deploy more agents, it becomes harder to manage them — let alone make them work together. Agent Connect is Dataiku’s answer to this challenge.
Agent Connect acts as a single conversational entry point where users can interact with any agent built in Dataiku. Behind the scenes, it handles routing, permissions, and collaboration between agents — even handing off tasks from one agent to another based on context or user role.
It ensures users don’t need to know which agent does what. They just ask — and the right agent responds. This is how you move from isolated agents to intelligent systems.
Connect to Any Model. Use the Right One for Every Job.
Behind every good agent is the right foundation model — and Dataiku doesn’t force you to choose just one.
With the Dataiku LLM Mesh, you can connect agents to leading providers like OpenAI, Anthropic, Mistral, and others. You can centrally manage access, route prompts based on use case, and easily test across models without rewriting your agents.
This flexibility makes it easy to keep up with rapid advancements in the LLM ecosystem. Want to adopt a newer version? Route certain workflows to a more cost-effective model? Restrict specific agents to an internal option? The Dataiku LLM Mesh gives you the abstraction layer you need to stay agile and in control.
Monitor, Debug, and Improve From Day One
Once agents are deployed, teams need a way to monitor their behavior, evaluate their performance, and make improvements — without guesswork.
That’s where Trace Explorer comes in. It provides a complete view into every decision an agent makes — including prompts, tool calls, inputs, and outputs — so teams can quickly understand what’s happening and why. When something goes wrong (or just feels off), Trace Explorer is how you fix it.
To ensure agents perform consistently over time, Quality Guard automates evaluation and monitoring using golden datasets, prompt scoring, and LLM-as-a-judge methods. Combined with Cost Guard, which tracks usage and enforces budget limits, these tools give IT and product teams everything they need to operationalize agents responsibly.
Governance Built In, Not Bolted On
For agents to move from experimentation to enterprise-wide adoption, governance can’t be an afterthought. It has to be integrated from the start.
With AI agents in Dataiku, governance is embedded at every layer. The GenAI Registry tracks every agent and LLM in use across your organization. Deployment sign-offs ensure nothing reaches production without proper review. And with value & risk monitoring, IT and compliance leaders can apply consistent qualification frameworks to agentic initiatives — just as they do with models and pipelines today.
This means every team can innovate freely, but always within a framework of accountability and trust.
Scale What Works, Stay in Control
With AI agents in Dataiku, you’re not just building agents — you’re creating an agentic foundation that scales with your business. From creation to orchestration, optimization to governance, every step is supported by one platform built for enterprise complexity and collaboration.
And because agents live alongside your analytics, data science, and machine learning workflows — not separate from them — they become a natural extension of how your organization works.