Skip to main content

Creatio AI Studio updates: October 22, 2026

With this update, you can decide how much an agent's model thinks before it answers. Agents can also read and post to the Creatio Feed with the new Creatio Feed skill. Scheduling gets easier to try out: run a new schedule once right after you create it, and see which schedules wait for a deployment. For knowledge sources, the Files tab shows how much text was read from each file and which files need attention. You can also set how long agents wait for the tools of an MCP integration. Learn more about all the new capabilities below.

Agent design and configuration​

Reasoning level for agent models. You can decide how much an agent's model thinks before it answers. In the agent's model settings, select a value in the Reasoning field. Lower values are faster and cheaper, and higher values are more careful. The list shows only the values that the selected model accepts, and the field appears only for models that accept a reasoning value. The "Model default" value keeps the model's own setting.

Skills​

Creatio Feed skill. Agents can work with the Creatio Feed of a record, a channel, or a single post. The new Creatio Feed vendor skill reads the feed. It also posts, comments, edits the user's own posts, and mentions people when the user asks for it. The Creatio integration has two new tools for this: one reads the feed and one posts a feed message. The read tool is added for every organization. When you ask AI Twin to build an agent that works with the feed, AI Twin adds Creatio Feed instructions to the agent. The feed tools are available when your Creatio connection supports the feed.

Knowledge​

Extraction details for each file. You can see how much of each file your agents can read. On the Files tab of a File knowledge source, the Characters, Chunks, and Pages columns show the figures for each file. Flags mark the files that agents cannot fully read, for example, "No text found" or "Not indexed." A counter above the table shows how many files need attention. Point to a flag to see what it means. File statuses stay the same.

Scheduling​

Test run right after you create a schedule. You can see what a new schedule produces without waiting for its first run. After you create a scheduled trigger, click Run once now and confirm, or click Skip. The agent runs right away if it is deployed in the schedule's environment. The current time must also be outside the schedule's quiet hours and excluded days. The run is a real run, so the agent can send messages or change records. To follow the run and see its output, open the agent's execution log.

Schedules waiting for deployment. You can see which schedules cannot run yet. When the agent is not deployed in the schedule's environment, an active schedule shows the "Waiting for deployment" status. The status appears on the agent's Triggers tab and in the Triggers section. When you set up such a schedule with AI Twin, AI Twin tells you that the schedule waits for the agent's deployment. AI Twin then offers to deploy the agent. If the agent stays undeployed through many scheduled runs, the schedule pauses. Resume it after you deploy the agent.

Limits for schedule values. Schedule values have limits.

  • The Repeat every field accepts up to 999 in any unit.
  • The Stop after failed runs field accepts a number from 1 to 10. Leave the field empty to never stop on failures. The default stays 5.
  • A schedule can start up to 1 year ahead and end up to 3 years ahead.

When you ask AI Twin for a value outside these limits, AI Twin asks you about the value instead of changing it.

Channels and integrations​

Tool execution timeout for MCP integrations. You can set how long agents wait for the tools of an MCP integration. On the integration's Configuration tab, enter the number of seconds in the Tool execution timeout (seconds) field. The value applies to every tool of the integration. Leave the field empty to use the default. The maximum value is set by the platform.

Agents that send from a shared mailbox. Agents listed in the Outbound senders block of an Email channel can send email from that mailbox. A listed agent can send after its next version is published. An agent that has its own mailbox keeps sending from that one. Replies to these emails go to the channel's inbound agent. A mailbox can have up to 50 senders. List each agent as a sender on one mailbox only. Listed agents follow the mailbox's Outbound recipients setting.

Web Chat channels that move between environments. After you switch a Web Chat channel to another environment, the widget on your website keeps working and uses the channel's saved settings.

Runtime and deployment​

Faster answers from agents with memory. Agents that use memory start answering sooner. The agent searches its memory with your message right at the start of the turn, without a separate planning step.

Memory recall for each agent. Each agent recalls the facts from your earlier conversations with that agent, even when you work with many agents. A long-term memory search for an agent covers only that agent's memories.

Observability​

Usage of realtime voice calls. Each turn of a realtime voice call records its usage on the call's run, so the run shows the tokens of the voice call.

More details in the run timeline. The model step of a realtime voice turn is named "Voice reply." When a prompt agent shows a widget in the chat, the widget call appears as a step.