Teradata has turned to open-source frameworks and platforms to offer a set of agent-building capabilities that it is marketing collectively as Agent Builder to help enterprises automate workflows.

These capabilities include a user interface, context intelligence, Teradata’s MCP Server and tools, and pre-built agents.

The user interface, which will be used to build agents and develop multi-agent workflows, is, however, not developed by Teradata and integrates third-party frameworks, platforms, and tools — an aberration from what rivals, such as Snowflake, Databricks, Salesforce, and ServiceNow, offer.

Currently, the choices of third-party frameworks and platforms include Flowise, which Workday acquired in August, and CrewAI, with support for LangChain and LangGraph to follow soon.

Other tools from Nvidia and cloud service providers, such as AWS, Microsoft, and Google, are expected to be added later, a company spokesperson said.

The Futurum Group’s lead for the CIO practice, Dion Hinchcliffe, calls Teradata’s integration approach, especially with open source frameworks and platforms, pragmatic, given the size of investment required to deliver robust agentic capabilities well.

“These frameworks already have robust communities and modular components, which let Teradata accelerate its time to market. This strategy allows Teradata to focus on its true differentiators: data, governance, and contextual knowledge,” Stephanie Walter, analyst at HyperFRAME Research, said.

Walter also pointed out that Teradata’s approach also avoids forcing customers to adopt a new proprietary system, unlike Salesforce or ServiceNow, which are building proprietary frameworks that lock customers into their ecosystems.

“This approach makes Teradata an agnostic infrastructure and knowledge provider, making it a trusted enterprise foundation for agentic AI,” Walter said.

MCP Server and pre-built agents included

The context intelligence capability inside Agent Builder, according to a company spokesperson, will include Teradata-developed expert prompts, contextual understanding, and tools, such as large language model (LLM) APIs.

These capabilities will help enterprises build agents grounded in their data, the company said.

The Agent Builder capabilities also include Teradata’s MCP Server, which enterprises can use to enable agents to carry out descriptive and predictive analytics via its Vantage platform.

Teradata’s MCP Server comprises database administrator tools, Enterprise Feature Store (EFS), vector store, RAG, metadata, SQL tools, a few select ClearScape Analytics tools, and some basic tools.

HyperFRAME Research’s Walter said the tight coupling with the MCP Server is “important.”

“MCP provides semantic access to enterprise data, curated prompts, and governance controls, which turns AgentBuilder into a data-driven, enterprise-grade framework, not just another LangChain wrapper,” Walter said.

Additionally, Teradata is offering some pre-built agents — Teradata SQL agent, Teradata data science agent, and Teradata monitoring agent — that enterprises can use as templates as part of Agent Builder.

While the SQL agent will help convert questions or requests expressed in natural language into SQL queries that can be executed against Teradata data warehouse tables, the data science agent can be used to create executable machine learning pipelines from natural language requests.

The monitoring agent, on the other hand, can be used to continuously monitor and manage Teradata databases, servers, and sub-systems, enabling optimization of system performance, the company said.

Analysts say that Teradata’s Agent Builder capabilities are highly likely to push enterprise CIOs to experiment with it.

“Enterprises that are experimenting with open-source agent frameworks but are struggling to productionize them as easily as they hoped will find Teradata’s value proposition appealing: trusted data, governance, hybrid deployment, and pre-built task agents,” Hinchcliffe said. Agent Builder is expected to be available in private preview by the end of the year.