Creatio AI Studio, October 22, 2026: Model Reasoning Level, the Creatio Feed Skill and Refined Schedules

The Creatio AI Studio update of October 22, 2026 is the third in October, after October 8 and October 15. It adds new model settings, a skill for working with the Creatio Feed, and several things that refine the schedules introduced the week before.

The full official notes: October 22, 2026 release notes. The notes do not mention feature flags, licensing requirements or beta labels. A general description of the product: Creatio AI Studio.

Reasoning level of the agent’s model

The agent’s model settings now have a Reasoning field. It sets how much the model “thinks” before answering: lower values give faster and cheaper answers, higher values more careful ones. The list shows only the values the selected model accepts, and the field itself appears only for models that support such a parameter. Model default keeps the model’s own setting.

What it gives you: simple tasks (classification, short replies) can run faster and cheaper, while complex ones can be set to a higher level, without changing the model.

The Creatio Feed skill

Agents can read and work with the Feed in Creatio: of a record, of a channel or of a single post. The skill can publish posts, comment, edit the user’s own posts and mention people when asked.

• The Creatio integration adds two tools: one to read the Feed and one to post a Feed message.
• The read tool is added for every organization.
• When AI Twin builds an agent that works with the Feed, it adds Feed instructions to the agent.
• Feed tools require a Creatio connection that supports the Feed.

Knowledge: file extraction details

On the Files tab of a File knowledge source, each file now shows Characters, Chunks and Pages columns. Flags such as “No text found” and “Not indexed” mark files the agents cannot fully read, and pointing at a flag explains it. A counter above the table shows how many files need attention. File statuses have not changed.

What it gives you: you can see at once which documents (for example scans without text) are not feeding the agent, instead of discovering it through wrong answers.

Schedules: refinements

• Run once now. After creating a schedule, you can click Run once now and confirm, or choose Skip. The agent runs immediately only if it is deployed in the schedule’s environment and the current time is outside quiet hours and excluded days. This is a real run: it can send messages or change records. You follow it in the agent’s execution log.
• “Waiting for deployment” status. An active schedule gets this status when the agent is not deployed in its environment (visible on the agent’s Triggers tab and in the Triggers section). AI Twin notes the dependency and offers to deploy the agent. If the agent stays undeployed through many scheduled runs, the schedule pauses and has to be resumed after deploying.
• Value limits: repeat every at most 999 in any unit, Stop after failed runs from 1 to 10 (empty means never, default 5), start up to one year ahead, end up to 3 years ahead. For out-of-range values, AI Twin asks instead of changing them.

Channels and integrations

• MCP tool execution timeout. On the integration’s Configuration tab, you set Tool execution timeout (seconds). The value applies to every tool of the integration, and an empty field means the default. The maximum is set by the platform (the notes do not state it).
• Outbound senders in the email channel. Agents added in the Outbound senders block of an Email channel can send from that mailbox after their next version is published. Agents with their own mailbox keep sending from it. Replies go to the channel’s inbound agent. A mailbox allows up to 50 senders, and an agent can be listed on only one mailbox. Listed agents follow the mailbox’s Outbound recipients setting.
• Web Chat channels across environments. After you switch a Web Chat channel to another environment, the widget on the website keeps working with the channel’s saved settings.

Runtime and observability

• Faster answers with memory. Agents that use memory begin answering sooner: they search memory using the user’s message at the start of the turn, with no separate planning step.
• Per-agent memory. Each agent recalls facts from earlier conversations with that agent, even when the user works with many agents. Long-term memory searches cover only that agent’s memories.
• Each turn of a realtime voice conversation records its usage on the call’s run, so the run shows the call’s tokens. The model step of a turn is named “Voice reply”, and when a prompt agent shows a widget in chat, the widget call appears as a step.

What to check after the update

1. Reasoning level. For agents that use the strongest model for simple tasks today, try a lower level.
2. Knowledge sources with files. Review the “No text found” and “Not indexed” flags.
3. Schedules. Before using Run once now, make sure the agent is deployed in the schedule’s environment and that you know what its run will do.
4. Email. Plan which agents send from which mailboxes (up to 50 per mailbox, one mailbox per agent).

Want to see how this fits your company?

As a certified Creatio Level IV partner, we help match agent settings (model, reasoning level, knowledge, channels) to specific tasks and keep their work under control. If you want to start with one use case, get in touch.

Other Creatio AI Studio updates

See also the other Creatio AI Studio updates from October 2026:

• Creatio AI Studio, October 15, 2026: agent schedules
• Creatio AI Studio, October 8, 2026: dictation in AI Twin, draft-first agent changes
• Creatio AI Studio: product overview, governance and rollout