Skip to main content

Creatio AI Studio updates: September 14, 2026

This update gives administrators more control over integrations, policies, and models. You can limit who can use an integration and test a PII protection policy in one environment first. New agents use GPT-5.6 Terra by default. Voice channels run the published agent version, and calls that start at the same time connect while capacity allows. The agent editor also logs draft edits and warns you when an environment runs an older release. Learn more about all the new capabilities below.

Agent design and configuration​

Draft edits in the audit log. The agent editor saves your draft one second after you stop typing. Draft edits appear on the Audit Log page as an "Agent updated" entry with the user's name. The entry shows that the draft changed, but not the changed values. It appears at most once every five minutes for each agent and user.

Model for memory search planning. Administrators can choose which model plans memory searches for their organization. Use Memory search planning on the Platform LLM page of Settings. If no model is selected, Use memory runtime default applies. If the selected model cannot be used, the agent's own model plans the search.

Models​

Default model and GPT-4o retirement. New agents use GPT-5.6 Terra, a Creatio-provided model, by default. Your organization can choose another model in Default model for new agents on the Platform LLM page of Settings. GPT-4o is retired and cannot be selected for new agents. Agents already set to GPT-4o keep running. Their model picker shows "The currently selected model is no longer available. Select another model from the list." If your organization's Default model for new agents is GPT-4o, new agents keep getting it until you change the setting.

Skills​

Larger skill files. Each file in a skill can be up to 512 KB, so longer instructions and reference documents fit. A file over the limit shows a warning when you run checks. It does not block work on the rest of the skill.

Knowledge​

"Me" in the Owner filter. In the Personal scope of the Agents, Skills, and Knowledge lists, the Owner filter always shows "Me" first, even when you own nothing yet. When you are the only owner, "Me" stays selected and the filter explains why. The owner list in Knowledge updates as soon as a knowledge source is created or deleted.

Channels and integrations​

Access control for integrations. You can limit who can use an integration and enable its tools. Open the Access tab of the integration and select Restrict access. Then add users, roles, or groups with Grant access. You can also limit each grant to specific access methods. A new integration that you create is available only to you and to users with the Manage security permission until you grant access to others. Integrations created before this update stay open to everyone until you restrict them. Integrations that the platform provides are always open to the whole organization. Runs without a signed-in user, such as triggered or channel runs, cannot use a restricted integration through a grant.

Voice calls on the published agent version. Calls to a voice channel run the agent version that is published and deployed to the channel's environment. Edits you have not published do not reach callers. A new voice channel is bound to an environment automatically. Once you bind an agent to the channel, the Routing block on the Overview tab shows the Environment, and you can change it there. For a voice channel created before this update, select its Environment in the same block to run the published version.

Concurrent voice calls. Voice calls that start at the same time connect as long as the platform has capacity for all of them. This works for calls by phone and from a web chat widget. If a web chat conversation was closed after a period of inactivity, a voice call started from it continues in a new conversation. The new conversation is linked to the earlier one.

Parameters for tools imported from an OpenAPI schema. When an agent calls a tool imported from an OpenAPI schema, it sends the query and path parameters that the schema declares. The external service gets the values that the agent supplied. If the service returns an error without a response body, the agent gets the HTTP status as an error it can report.

Trust and governance​

Per-environment activation for PII protection policies. You can test a PII protection policy in one environment before it applies elsewhere. Use the Active environments card of the policy on the Policies page of the Trust & Governance section. Select Add environment to limit the policy to the environments you choose. A policy with no environments selected shows Active in all environments. If the environment of a run is unknown, a policy limited to specific environments still applies to that run.

Runtime and deployment​

Deploy-required notice for older releases. The Skills tab of an agent shows Deployment update required when an environment runs an older release than the latest one you published. This can happen, for example, after you deploy an earlier release. The notice names each environment that runs an older release. It clears once every active environment runs the latest published release. The notice updates when you deploy or roll back the agent. An agent that already ran an older release shows it after its next deployment.

Observability​

Memory steps in the run timeline. In the run details on the Agent Observability page, each memory step appears inside the iteration that triggered it. This shows which step of the run used memory. Runs recorded before this update keep their old layout.

Usage and budgets​

AI credits as the default usage unit. Usage and limits are measured in AI credits, the unit that reflects how AI usage is priced. Usage analytics, agent metrics, and budget limits all use AI credits. The Usage unit setting in Settings shows AI Credits (default) as the recommended option.