AI Engine Help Center

Release notes

Notable changes in recent AI Engine versions that users and admins will see, and what to check after upgrading.

This page summarises the notable user-visible changes in recent versions of AI Engine. Your org gets a change when it is upgraded to a version that includes it. To see which version you have, go to Setup → Installed Packages.

Sales Agent chat#

  • Live progress. While the chat works, it shows what it is doing, for example "Understanding your question", "Searching Opportunities", "Reading 42 opportunities" and "Writing the answer".
  • Answers in the language of the question. The chat replies in the language each question is asked in, and you can switch language between questions.
  • Larger result sets. Show all loads up to 2,000 rows in the panel, and CSV and Excel export cover the full result set up to 10,000 rows. When a question matches more records than the table shows, the answer gives the total.
  • Activity Digest. General questions such as "what happened this week?" return a summary of new and changed records across the objects enabled for the digest.
  • Modern Interface. An optional chat theme, switched on in the Org Profile. The classic theme remains the default.
  • Company News as cards. News results appear as cards with the date, headline, publisher, a one-sentence summary of why the news matters, and a link to the source.

Prompt templates#

  • Scope. A template is either Record (runs on a record, optionally only on selected objects) or Global (appears in the Actions menu when the chat is opened without a record, for example in the utility bar or on an app page).
  • Data sources. Up to five active saved queries per template add live CRM data to the prompt through {data:name} placeholders. Queries can be generated from a plain-language question, previewed exactly as the AI will receive them, and filtered to the running user or the current record. Each data source's Row limit is used at run time, and Preview shows the effective limit. See Prompt data sources.

Models and providers#

  • Fast model. An optional cheaper model from the same provider for web-search steps and Help mode. See Supported providers and models.
  • Model list in the Org Profile. OpenAI and Anthropic models are chosen from a list with short notes on each.
  • Newer models. Newer Claude models are in the model list (some marked experimental there), and Azure OpenAI deployments of reasoning models (o-series and the GPT-5 family) are supported. AI Engine recognises these by the Azure deployment name, so name the deployment so that it starts with o1, o3 or o4, or contains gpt-5.
  • Meet Prep web search is optional. Adding web news to meeting briefs is controlled by Meet Prep: include web search in the Org Profile. It is off by default.

AI Scoring#

  • Scoring Context Fields decide exactly which fields the AI sees when scoring.
  • Scoring Instructions let you add your own business rules and label names, and Output Language sets the language of the written reason.
  • Verify Company On The Web optionally checks each scored company with one web search.
  • Re-score already scored records lets the batch button refresh existing scores.

Activity Monitor#

  • Every provider call is counted. For actions that make several calls, the AI Activity Log record holds the tokens of all calls, the number of calls, cached tokens, web searches and a per-model breakdown. These details are on the log records and in reports; the Activity Monitor dashboard shows the totals.
  • Cost estimates are calculated per call with the model actually used, with cached input priced at a discount. Actions that make several calls show the total of all of them.

Help mode#

  • The ? Help mode in the chat answers from a built-in help guide for users and admins, and you can add your own topics. See Help extension.

After upgrading#

  • Meet Prep web search: if your meeting briefs included web news before, switch on Meet Prep: include web search in the Org Profile.
  • Cloned permission sets: new fields, objects and permissions are added to the packaged permission sets only. If your users have clones, add them there by hand.
  • Sharing settings: an upgrade does not change your org-wide defaults. In Setup → Sharing Settings, check that Default Internal Access for AI Activity Log is Private; see Recommended security settings.
  • Old screens or "Unable to find Apex action method": do a hard refresh of the browser page (reload without cache), or sign out and back in.

More:

Still stuck? Ask AI, or email contact@cloudconnected.pl with the exact error message, the page or action you used and roughly when it happened.