Skip to main content

Creatio AI Studio updates: August 2026

In August 2026, Creatio AI Studio received updates across the product. They include AI Twin, agent design, channels and integrations, trust and governance, and observability. Learn more about all the new capabilities below.

AI Twin​

Answers about closed deals and reporting periods. AI Twin can answer questions about deals won or lost in a period. It can also tell you which stage a deal was in when it closed, based on the deal's stage history. For a period such as "Q2" or "this fiscal year," AI Twin checks the fiscal year settings in Creatio. It then tells you which dates it used. If no fiscal year is set up, AI Twin asks which period you mean. A group or segment you name is matched to its saved folder.

Exact agent identification. AI Twin works with an agent only when the name you give matches exactly. If nothing matches, AI Twin says it did not find the agent and offers the list of agents or a search by name. Agent lists in AI Twin show each agent's name, status, and scope.

Agent card after an edit. When AI Twin changes an agent, it shows the card of that agent with its actions.

Ordered tool confirmations. When an agent asks you to confirm an action in AI Twin, the request appears before the answer it is about. The order stays the same after you reload the page. The request names the agent that is waiting for you, and your confirmation is acknowledged once.

Stable conversation history. Every message in an AI Twin conversation stays in place and in time order. An answer you stopped is still there after you reload the page, and your next message comes after it. A finished answer shows its text once.

Agent design and configuration​

Single run time budget for agents. Each agent has one time limit for its runs: Run time budget (seconds) in the Runtime limits block of the Advanced tab. It applies to every run, whether a channel, the test chat, or an API call starts it. The separate timeout field is removed. Agents that used only that field keep the same run time.

Inline prompts on every channel. A prompt agent uses the prompt in its System Instructions in every channel and in evaluations. The deployed agent behaves as it does in the test chat.

Agent designer in multiple tabs. You can work in multiple agent designer tabs at once. When you save, publish, or deploy in one tab, the other tabs stay up to date without a reload. If you open four or more designer tabs at the same moment, you may briefly reach the request limit for your user.

Owner filter in the Agents, Skills, and Knowledge lists. In the Personal scope of the Agents, Skills, and Knowledge lists, the Owner filter lists every owner, even when there is only one. If there are no owners, the filter reads "No owners." An agent card without a description reads "No description."

Models​

Built-in tools follow the selected model. If you switch an agent to a model that cannot run some of its built-in tools, the editor turns those tools off. A notice names them. A new agent starts only with the built-in tools its model supports. You cannot save an agent with a built-in tool that its model cannot run.

Creatio-provided model group. In model pickers, the Creatio provided group lists only the models that Creatio provides. A model your organization serves with its own key is listed under its vendor. The group stays available even when another model service is down.

Skills​

Skill version numbering. The latest version number of a skill always matches the newest saved version. This holds whether you edit the skill or save, restore, or activate a version from its list. A skill stays editable no matter how many versions it has.

Vendor skill versions. Skills from the vendor registry show the vendor version next to your version number, for example, "Version 2 · vendor v5" in the version selector. You see it in the skills list, the version selector, the Versions tab, and the agent's Skills tab. When you update a vendor skill, every vendor version in between is added, and the earlier versions are listed as archived.

Reporting periods skill. The Resolve reporting periods vendor skill turns periods such as "this quarter," "last fiscal year," or "FY27" into exact date ranges. It uses a definition you give or your organization's fiscal settings. If neither is available, it asks you. Import it from the vendor registry in the Skills library.

Knowledge​

Search inside attached files. An agent can find a phrase in a file attached to the conversation and quote the matching lines. It does not need to read the whole document, so it answers questions about a specific value or passage directly. Search ignores letter case. Images cannot be searched.

Long attached files read in full. An agent reads a long attached file page by page, so it can use content from any part of the document.

Agent evaluation​

Evaluation runs that match chat behavior. An evaluation tests the agent as it behaves in a conversation. Each run gets the same instructions, temperature, and environment as the agent in chat. The agent calls its tools during a case, and a case that depends on a tool is graded on the tool's real result. For these cases, run the evaluation in an environment where the agent's integrations have credentials.

Evaluation environment on every run. The Evals tab shows the environment of each evaluation run, using your environment names. The environment you select stays selected when you switch between runs and cases, reopen a dialog, or reload the page. Re-running an evaluation uses the environment of the original run. A run recorded before environments were tracked shows "Unknown environment" and asks you to choose one first.

Tool diagnostics in evaluation runs. The details of an evaluation run list the tools the agent was granted but could not call, for example, because an integration has no credentials. Each case that expects one of those tools is marked "tool unavailable." The case status still shows the graders' verdict. Credentials are checked for the whole organization, not per environment.

Channels and integrations​

Agent instructions on voice calls. A voice call follows the same base instructions as the agent's chat conversations, for every kind of agent. Topic, policy, and safety rules apply on the phone too. The Voice system prompt in the Voice profile block changes only how the agent speaks, such as tone, length, style, and pace.

Creatio tools on voice calls. Voice agents can use the tools of Creatio integrations during a call, for example, to create a record. This also works for calls started from a web chat widget. Each tool call uses the environment of the voice channel, so tool access works the same way as in chat. The voice channel must be bound to an environment.

Telephony diagnostics for Twilio. The Diagnostics page of the Telephony section shows a working Twilio integration as configured. The Configuration checklist covers Twilio credentials, Public Base URL, and At least one phone number. The Integration health card shows Outbound ready when the integration can place calls. The Voice webhook URL under Inbound setup & verify matches the address on the Phone numbers page.

Credential status for delegated integrations. A tool that uses an integration with delegated access is shown as available when the integration has an enabled binding. The credential comes from the signed-in user when the agent runs. You see this status in the tool list of a run and in the evaluation tool diagnostics.

Longer voice preview calls. A voice preview call in the browser can last up to 30 minutes, the same as a regular voice call.

Searchable agent pickers on email and web chat channels. You can find an agent by typing part of its name in the agent pickers of email and web chat channels. This helps when many agents have similar names. On an email channel, search works when you choose the inbound agent or allow another agent to send. On a web chat channel, it works when you bind an agent. When not all agents are loaded, the picker says that it shows only the first ones.

Trust and governance​

Permission limits for the Individual Contributor role. The Roles tab of the Security section shows which permissions the Individual Contributor role cannot have. In the Add permissions dialog, these permissions are marked as not available for this role: Edit agents, Edit agent configuration, Create deployments, Promote deployments, Edit channels, and Route channel traffic. A role saved with one of them is refused. To grant these permissions, create a custom role.

Chat access for the AI Chat User role. Users with only the AI Chat User role can open and use the chat. The role includes the Read channels permission that the chat needs. It still cannot edit channels or create agents, including personal agents. Existing organizations get the permission automatically.

Sign-out and session renewal. Sign out ends your Creatio AI Studio session and opens the sign-in page. When a session expires and cannot be renewed, Creatio AI Studio takes you to the sign-in page right away.

Fiscal period labels preserved by PII protection. PII protection keeps fiscal period labels intact. An agent that reports a period such as "FY2026-2027" returns the full label. A phone number next to such a label is still masked. Custom detection rules do not change.

Read-only access to other users' personal resources. With the View others' personal resources permission, a role can view other users' personal agents, channels, knowledge sources, and skills, including their runs and logs. It cannot edit them. With Manage others' personal resources, a role can also edit the resources it can view. You set both on the Roles tab of the Security section.

Runtime and deployment​

Date handling that follows the request. When you ask an agent to schedule or create something without a date, it picks the nearest future date. When you ask about existing data, it uses the period you asked about, including past periods, and tells you which dates it used. The current date is taken in UTC.

Observability​

Agent version in run history. You can see which version of an agent produced each run. The Agent Observability page shows the Agent version next to the environment of every run. The run details show it too.

Clearer run details. The run details on the Agent Observability page show more about each run. Each model call lists the Available tools sent with the request, up to 100 tools. The details also show the agent's prompt. This is the current prompt, so it can differ from the one an earlier run used. Each tool call in the timeline has its own outcome, so a failed tool call shows as failed even in a completed run. Some runs are started by people who are not Creatio AI Studio users, such as web chat visitors. For these runs, Initiator reads "External participant" and a short ID.

Usage and budgets​

Remaining quota on the Usage analytics page. The AI Credits remaining card on the Usage analytics page shows how much of your organization's subscription quota is left. It updates each time the page loads. With another usage unit, the card reads AI Actions remaining or Tokens remaining.

Lower token use for skills and MCP tools. Runs that use skills or get structured results from MCP tools use fewer tokens.

Limit refusals in conversations and run history. When a usage limit stops a run, the person who sent the message sees a limit message. This works in AI Twin and in channels such as web chat. In AI Twin, the message stays in the conversation after you reload the page. In channels, a limit reached partway through a run shows the same message as a limit reached before the run. For API requests, the refusal includes a clear sentence that names the limit. Each refused run is recorded with the Failed status on the Agent Observability page, even if it was refused before it started.