AgentFlow is Reltio’s conversational interface for working with your master data. Built on the Model Context Protocol (MCP) server, it allows you to explore, validate, and manage entity records through chat-based interactions with AI agents. These agents operate on live production data while honoring your organization’s role-based access control (RBAC), data masking, and audit policies.

AgentFlow is designed for the following business and technical users who need to manage and govern master data in real time:

Data Steward Solution ArchitectSystem Administrator Data Product Owner

For more information about each of these user roles, see About roles.

How it works

You launch a conversation with a selected agent in the AgentFlow workspace. Agents use governed APIs (called tools) to read and act on entity data. Each interaction — whether it's a lookup, enrichment request, or merge action — is scoped to your access level and fully auditable.

Agent types

AgentFlow supports two categories of agents:

• Included agents like the Data Explorer Agent let you look up entities, explore relationships, and summarize data. These are read-only and available to all AgentFlow users.
• Licensed agents such as the Match Resolver Agent let you take action on entity records — such as merging or rejecting duplicates — and may also perform enrichment from public sources.

Core concepts

AgentFlow relies on a few key concepts that shape how you interact with your data:

• Conversation: A chat session between you and one agent, scoped to a specific task. The conversation retains context as you continue interacting.
• Agent: An AI assistant with a focused skill set. Each agent uses approved tools to read, compare, or update data — depending on your permissions.
• Tool: A secure, governed capability the agent uses behind the scenes to complete an operation (for example: search, match, merge). Every call is logged and access-controlled.

Built-in safety and compliance

Every action an agent performs is routed through the MCP server. This ensures:

• All actions respect your RBAC policies and attribute-level masking
• Write operations (e.g., merges) are logged with timestamps and user attribution
• Agents can only use allowlisted tools and cannot access raw databases
• Agent behavior is constrained to your tenant's policies and data residency zones

When to use it

To begin using the interface, see Use the AgentFlow workspace.