AI Agents. Enterprise-Grade. Campaign-Ready.

Build custom agents using natural language

Treasure Data includes a no-code/low-code interface as a part of the AI Agent Foundry where users can build, test, and manage AI agents in natural language. Even non-technical users can design AI workflows that fit their teams without needing deep LLM engineering expertise.

Request a consultation with our AI and Personalization Services group to start your custom agent journey.

How it works:

• Users define prompts, tasks, and objectives for agents using plain English
• Agents can be powered by top-shelf LLMs and connected to Treasure Data’s query engines, pipelines, and customer data
• Specify input formats, output expectations (e.g., SQL, segment, text), and even embed tools the agent can call
• Customizable guardrails ensure agents stay on track with business needs and compliance

Access a catalog of out-of-the-box agents

The Foundry includes a growing catalog of ready-to-use agents like Audience Agent, which are prebuilt, optimized and included out of the box, giving you immediate value without needing to “start from scratch.”

With a click of a button, you get access to Audience Agent and any future product agent within your CDP. Agents can be extended later via custom development.

How it works:

• Audience Agent can be used instantly for audience discovery, micro-segmentation, and campaign planning using chat
• All Agents are powered by the Foundry, which means you get the power of the Foundry even if you don’t build custom agents right away
• Other agents include Journey Optimizer Agent, Deep Research Agent, Content Generator Agent, etc.

Ground agents with a built-in Diamond Record

While other platforms build agents on flaky or partial data, every agent in the AI Agent Foundry is automatically grounded in clean, unified customer data from the Diamond Record, which is Treasure Data’s trusted source of truth. This ensures every response, recommendation, or action is rooted in accuracy—a must for enterprise use.

How it works:

• AI agents reference the same ID-resolved, privacy-compliant dataset used across the CDP
• Define which attributes, segments, or tables an agent has access to
• Results are always based on high-quality, governed data, not hallucinated or out-of-context guesses

Ensure secure, scalable agent governance

Security, auditability and operational oversight are critical for enterprise AI agent adoption. The Foundry bakes this in from day one, providing an enterprise-grade framework for managing who can build, access and run agents, with full auditability and compliance support.

How it works:

• Role-based access controls govern which users or roles can interact with different agents or data sets
• Admins can track usage by team, agent and conversation count
• Guardrails can be set to limit agents to certain datasets or outputs (e.g., read-only vs. edit access)
• Foundry also supports telemetry and lifecycle monitoring for performance and health tracking

Build templates and reusable agents

You don’t need to reinvent the wheel every time you want a new agent. Agent Blueprinting is a system for templating and reusing agents across teams, regions, or brands, supporting repeatable deployment at scale. Blueprinting accelerates time to value and enables the reuse of best practices.

How it works:

• Successful agents (e.g., for promotions, churn prevention, product recommendation) can be cloned or parameterized
• Blueprints save prompt instructions, tools, data access scopes, and execution modes
• Deploy these blueprints to new teams or geos in minutes instead of weeks

Design multi-agent orchestration

Agents can call or delegate tasks to other agents by creating a workflow of intelligent actions across multiple tools or touchpoints. This is where AI becomes truly autonomous, mirroring how humans work in teams with distributed tasks, logic, and execution.

How it works:

• Define a primary agent (e.g., a campaign planner) that calls a helper agent (e.g., a segment builder or content writer)
• Each agent performs its task and passes results forward
• Orchestration logic can be built into prompts