Our most recent community show featured a deep dive into Reltio AgentFlow™ Agent Builder, the newest capability enabling customers to design, test, and deploy custom AI agents grounded in governed, trusted data. The session, led by Aleem Javed, walked through the full agent lifecycle — from authoring to publishing — and offered a sneak peek at what's coming next.
What Is Agent Builder?
Agent Builder is Reltio's no-code solution for building custom AI agents on top of your trusted Reltio data. Until now, AgentFlow users could only interact with pre-built agents such as the DEA agent or the Merge Resolution agent. Agent Builder changes that, giving teams the ability to define an agent's instructions, tools and capabilities directly in the UI and test it in draft mode before anything goes live — no developer experience required.
Governance Built In from Day One
One of the central themes of the session was enterprise-grade governance. Agent Builder enforces a maker-checker model: the person who builds an agent cannot be the same person who approves it. Three roles govern the process — Agent Author, Agent Approver, and Agent Admin — and the separation between author and approver is enforced by the platform, not just by policy. Agents move through clearly defined states (draft, pending review, published, archived), every step is timestamped, and a full audit trail is available at all times.
The End-to-End Agent Lifecycle
Aleem walked through the complete flow live:
-
Author: Define the agent's name, instructions (system prompt), capabilities, and the tools it can access from Reltio's MCP server — which offers 100+ tools across search, entity management, merge, and more
-
Test in draft mode: Run real conversations with the agent before submitting, including an execution log showing credit consumption and tool calls (visible only in builder mode)
-
Submit for review: Once submitted, the agent definition is locked; safety checks run automatically for policy violations, prompt injection attempts, and inappropriate content
-
Approve and publish: The approver reviews the full prompt, tool list, and safety scan results, and can test the agent themselves before approving — at which point it becomes available to all users in the tenant
Agents are also portable: published agents can be exported as an encrypted .agent file and uploaded to another tenant, where all details pre-fill automatically and the same approval flow applies.
Coming Soon: Agent Blueprint
Aleem previewed Agent Blueprint, currently in internal testing, which directly addresses the "blank page problem" — the challenge of not knowing where to start when building a new agent. Users simply provide a natural language description of what they want the agent to do, and Agent Blueprint generates a structured system prompt following Reltio's best-practice guidelines, recommends tools with reasoning, and auto-selects them in the next step. What might take several minutes of prompt crafting gets done in one to two minutes, and the result is fully editable.
Where Custom Agents Create Value
Aleem outlined four patterns where custom agents deliver the most impact today:
-
Data quality: Catching duplicates as records come in and suggesting merges before issues spread
-
Record understanding: Letting sales, service, or care teams ask an agent to explain a profile rather than clicking through screens
-
Compliance and risk: Running KYC and onboarding checks against source-backed data that can withstand scrutiny
-
Operational workflows: Agents that read trusted context and take specific approved actions inside a defined process
The direction of the platform is toward greater autonomy — agents handling large batches of records independently and routing only ambiguous cases to a human reviewer.
AgentFlow: The Bigger Picture
Agent Builder is one capability within the broader AgentFlow platform, which also includes pre-built agents, Unstructured Data Studio, and data quality features. What ties everything together is that all agents run on governed Reltio data — unified (one profile per entity), governed, connected through the Intelligent Data Graph, and delivered in real time rather than from stale extracts. Aleem described this as a "system of context for AI": the same model, given better context, produces more accurate and more trustworthy results.
Q&A Highlights
The session included a lively Q&A. Key topics covered:
-
External MCP servers: Support for custom MCP server onboarding is on the roadmap
-
Prompt templates: Agent Blueprint will handle this; guidelines are also available on the Doc Portal, and agents will become cloneable as a starting point
-
Out-of-the-box agents: Available in the Discover Agent section of AgentFlow (Resolver, Unmerger, Data Explorer, Work Assigner, Address Signature, Profiler, and more)
-
Consistency: Custom agents built for a specific organization's data model will outperform general-purpose pre-built agents for that use case
-
RBAC: Agents currently run under the calling user's access permissions, so data-level security is already respected; agent-level access controls (restricting which users can see which agents) are on the roadmap
-
Search and match merge in one agent: Fully supported through the MCP server's read/write tools
What's Coming Next
Near-term roadmap items include Agent Blueprint GA, email notifications for critical workflow actions, and agent cloning. Larger investments underway include agent memory (retaining decisions across sessions), long-running task execution (autonomous batch processing), file upload support, knowledge base grounding (bring your own documents), built-in evaluation tooling, and the ability to promote agents across tenants.
The recording and any follow-up resources will be posted to the community within the next few days. If you have additional questions that weren't addressed during the session, feel free to post them in the community.
#Featured
#CommunityWebinar
#AIML