Version 5.5

Version 5.5

10 August, 2026

AI Agents

• Improving an Existing AI Agent with AI:
  An existing AI Agent can now be improved by chatting with an assistant instead of editing its fields by hand, with no test run needed first. An "Improve with AI" button on each agent card on the AI Agents page opens Agent Improver, which proposes its changes as a list to approve before anything is applied. It is available on the Draft version only, and approved changes are applied as a single batch.

• Generating Test Scenarios with AI:
  Test scenarios can now be generated from the agents already configured in a project, so testing can start without writing each one by hand. The "Generate with AI" button on the AI Testing page opens a modal where the agents to base the scenarios on are chosen, together with optional notes. The results are reviewed and selected before they are added, and they are saved rather than run.

• Custom Attributes:

  • Setting Custom Attributes in a Flow:
    Values already known inside a flow can now be written to a conversation attribute without an LLM call, so data such as a live agent transfer is available for reporting. They are defined on the new "Custom Attributes" tab of the "Conversation Attributes" section in Project Settings, and a "Set Attribute" node sets their value in any flow. A project can hold up to 20 attributes, and a key and output type are fixed once created.
  • Custom Attributes on the Conversations and Dataset Pages:
    Custom attribute values are now visible on the Conversations pages and can be filtered on, so what a flow recorded can be analysed. They can be added as columns on the conversation list, are included in the export, and a "Custom Attributes" filter section sits below "AI Insight Filters" on both the Conversations and Dataset pages. Nothing is shown or exported when Conversation Attributes is switched off for the project.

• Data Tables:

  • Inserting a Row from a Data Table Node:
    A Data Table node can now add rows to a table as well as read from it, so a flow can record what it collects. An "Insert" operation in the node's "Operation" list opens a "Values to Insert" section where each column is given a static value or a variable. Insert adds a single row, and columns left empty are saved as null.
  • Returning Only the Number of Matching Rows:
    A Get operation can now return just the number of matching rows instead of the rows themselves, so a flow can branch on whether records exist without the cost of fetching them. A "Return count only" toggle sits above "Response Content" and is off by default, so existing Get behaviour is unchanged.

• Deleting Execution Details on Their Own Schedule:
  Execution details (tool calls, guardrail results, HTTP traces and invoke failures) can now be deleted sooner than the conversations they belong to, since they are usually only needed shortly after a conversation ends. The period is set under "Data Retention" in tenant features, alongside session retention, and must be shorter than it. Once a conversation's execution details have expired the detail screen says so, and they cannot be recovered.

• Adding Your Own Instructions to ACR:
  Your own instructions can now be given to Advanced Conversation Routing, so business rules and tone or language preferences shape how it talks to customers. An optional "Additional ACR Instructions" field appears in the advanced settings of a Virtual Agent node once ACR is enabled. What is entered is appended to the default system prompt rather than replacing it, so standard intent classification is unaffected.

Improvements

AI Agents

• AI Insights:

  • Ordering AI Insights:
    The order AI insights appear in can now be chosen, so the ones that matter most come first. Each insight card on the AI Insights settings page has a drag handle, and the order set there is used on the conversation detail screen, the Conversations page and the dataset filters. Ordering is done on the Draft version and takes effect when the project is published.
  • Pre-filled Ranges on Score Insight Filters:
    A score insight filter now opens with the minimum and maximum from the insight's own definition already filled in, so the range no longer has to be typed each time. A value entered outside that range is corrected back to the nearest allowed one.

• Preserving Complex Tables in Document Processing:
  Documents added to a Knowledge Base now keep complex tables intact, so their structure survives ingestion. The default Intelligent Document Processing prompt renders tables with merged cells or nested headers as HTML, while simple tables stay in markdown. This applies to knowledge bases created from now on; an existing knowledge base keeps its current prompt.

• Filtering the HTTP Metrics Dashboard by Endpoint:
  The "HTTP Metrics" dashboard can now be filtered by endpoint, so one call can be examined without the others in the way. The status code widget has also moved higher on the page, and chart averages are drawn as dashed lines.

• Remembering Execution Details Settings:
  The execution details options chosen on the conversation detail and Test Conversation screens are now remembered for each user within a project, so the same toggles do not have to be set again on every visit.

• Reading Instruction Comparisons More Comfortably:
  Long instruction lines can now be read without scrolling sideways when two agent instruction versions are compared. The compare window has a word wrap option and a maximize control that expands it to near full screen, and it opens at its default size again each time.

• Minor UI Improvements:
  Several minor improvements have been made to the user interface to enhance user experience and usability.

• Minor Bug Fixes:
  Various minor bugs have been addressed to improve overall system stability and performance.