Dataiku, The Universal AI Platform, has launched AI Agents with Dataiku, a new set of capabilities designed to create and control AI agents at scale.
"AI is raw power — and it's time for companies to take control. Companies are on the verge of repurposing two decades of enterprise applications built on systems like Snowflake, Workday, and SAP with a new layer of AI-native applications. These applications demand a combination of analytics, models, and agents that only Dataiku can deliver," said Florian Douetteau, co-founder and CEO of Dataiku.
Centralised agent creation
Dataiku offers a range of options to cater to different user needs. It supports the central creation of agents with a visual agent, a no-code option ideal for non-technical business users, and a code agent, a full-code option suitable for developers. Its key capabilities include Managed Agent Tools, which ensure the quality and validation of agents' tools; GenAI Registry, for strategic oversight; and Sign-offs for risk monitoring.
Dataiku LLM mesh architecture
Dataiku is committed to providing IT with the tools to orchestrate agents at scale. The Dataiku LLM Mesh is a key player in this, simplifying the management of model access across all providers. Together with Dataiku Safe Guard and Agent Connect, this forms a powerful trio that centralises agent access across the organisation from a single interface.
Observability and performance monitoring
Dataiku also provides agent observability and performance monitoring capabilities, ensuring full transparency. This includes Trace Explorer, which offers a complete insight into agent decision-making, input/output flows, and debugging; Quality Guard, which evaluates and monitors agent performance; and Cost Guard, which tracks usage in real-time, enforces budgets, and handles internal rebilling.
Connecting agents to business
Dataiku claims to support all major cloud environments, model providers, and data platforms for agents to run wherever enterprise data and analytics already live. It can fully integrate agents into existing data pipelines, MLOps workflows, and model governance processes.
