Creatio adds memory search planning and observability to AI Studio

Creatio released a September 14, 2026 AI Studio update that introduces configurable memory-search planning and enhanced observability for agent memory use, giving administrators and teams finer control over how agents access and consume stored data Creatio.

Memory-Search Planning Now Configurable

Administrators can now select which large language model (LLM) plans memory searches on the Platform LLM settings page Creatio. If no model is selected, the system defaults to the memory runtime's built-in model; if the selected model becomes unavailable, the agent's own model takes over planning. This layered fallback approach prevents agent workflows from breaking when a preferred model goes offline.

The feature addresses a core operational challenge in agentic workflows: memory searches can consume significant API calls and latency if poorly planned. By centralizing model selection at the platform level, Creatio lets enterprises standardize which reasoning engine performs these queries—allowing teams to balance cost, speed, and accuracy across all agents rather than configuring each agent individually.

Memory Steps Now Visible in Run Timelines

The update adds a new Agent Observability page feature: each memory step now appears inside the iteration that triggered it, showing precisely which step of an agent run accessed memory and when Creatio. This granular visibility helps operators debug agent behavior and audit memory usage without reconstructing call logs manually.

Runs recorded before the September update retain the previous layout, ensuring backward compatibility for historical audit trails. New runs automatically display the improved structure.

New Permission: Read Run Step Content

Creatio's September 23, 2026 release notes introduce a separate "Read run step content" permission that controls whether users can view the actual data within observability steps for runs initiated by other team members Creatio. Without this permission, users can still see each step's execution count, duration, and status—useful for monitoring throughput—but cannot access the step's content, protecting sensitive data or customer information.

This permission model lets enterprises grant broad visibility (step counts and performance metrics) to operations teams while restricting detailed content access to security or compliance officers, addressing a common challenge in regulated industries where multiple stakeholders need observability without universal data access.

Why This Matters for Enterprise Agents

As enterprises deploy agents at scale, memory management becomes a critical control point: decisions about which model searches stored context affect cost, latency, and consistency. The ability to audit memory usage step-by-step, combined with granular permissions, moves agent workflows closer to the observability standards required in financial services, healthcare, and government sectors.