Documentation Index

Fetch the complete documentation index at: https://docs.knovvu.com/llms.txt

Use this file to discover all available pages before exploring further.

Version 5.4

  • Published on Jul 20, 2026
  • 6 minute(s) read
  • Suara Atay
 Prev Next

Version 5.4

20 July, 2026

AI Agents

  • Data Table:

    • Data Tables Import CSV:
      A data table can now be created by uploading a CSV file instead of defining columns manually, making it easy to bring existing data onto the platform. After upload, a preview modal shows delimiter and header detection, per-column type inference, and a data sample, letting users adjust the delimiter, toggle the header row, and review results before import. Once imported, columns can still be renamed, retyped, or added up to the 20-column limit, with the same naming and system-column validations used in manual table creation.
    • Data Tables - Fast Filter Columns (Indexing):
      Columns can now be marked as "Fast filter" when creating or editing a Data Table , enabling indexed, faster retrieval when that column is used as a filter condition in the Data Table node. Up to 5 fast filter columns are allowed per table, available for Text, Number, Boolean, and DateTime types (Long Text is excluded).
    • Data Table Node - Sorting:
      A new optional "Sort by" field has been added to the Data Table node's Get operation, allowing returned rows to be ordered by a selected column in ascending or descending order before the row limit is applied. This enables deterministic results — such as the most recent record or the highest/lowest value — for consistent branching logic in flows. Sorting respects each column's data type (numeric, chronological, or lexical), uses DateCreated as a tiebreaker, and handles nulls by placing them first in ascending order and last in descending order; when left empty, rows are returned in their default order.

  • AI Agent Builder:
    A "Build with AI" option is now available on the AI Agent create card, letting scenario designers describe the agent they need in a chat panel and have it built automatically. The Builder Agent gathers requirements through guided prompts, then creates either a single agent or a supervisor with its sub-agents using the LLM configured under the new "Agent Builder" tab in Project Settings. Once creation completes, a visual summary of the resulting agent structure is shown with a "View Agent Details" button that navigates directly to the new agent's (or supervisor's) detail screen.

  • AI Insights:

    • Custom AI Insights:
      The AI Insights tab now supports user-defined insights in addition to the prebuilt ones, letting teams extract custom data points from conversations for reporting. Each custom insight is defined with a name, description, and output type (single or multiple values — category, free text, or numeric score with configurable range or enum values), can be toggled on/off without deletion, and is limited to 20 per project with duplicate-name prevention. Once created, an insight's name and output type are locked, though descriptions and possible values remain editable, and enabled custom insights are exposed to post-session flows through a new "InsightsCustom" variable for use in downstream logic.
    • Showing AI Insights on Conversations Page:
      AI Insight results (both prebuilt and custom) are now surfaced throughout the Conversations pages. They can be added as columns and included in exports on the Conversations List, filtered through a new dedicated "AI Insights" filter, and viewed in full on the conversation detail screen under a scrollable "AI Insights" section.
    • Add AI Insights to Dataset Creation:
      Datasets used for AI Testing can now be narrowed down with a new "AI Insight Filters" section, letting users target specific customer conversations by insight value when building a sample. Each filter row lets an insight be selected from the project's draft version, with the operator and value inputs.

  • HTTP Node Metrics Dashboard:
    A new "HTTP Metrics" tab has been added to the dashboard, giving designers and support teams visibility into HTTP node request performance across a flow. It surfaces summary widgets (total requests, total errors, error rate, average duration), timeline views for request volume and response duration (average, P95, P99), pie charts for sync/async and status code distribution, and a per-endpoint breakdown table with method, totals, error rate, and average duration, all filterable by channel and date.

  • Showing Guardrail Details in Test Chat & Conversations Page:
    Guardrail evaluation details can now be inspected directly in Test Chat and the Conversations page, showing the masked input sent to the LLM and the unmasked output returned. A new "Show guardrail details" setting has been added under Monitoring in Test Chat, which displays "Guardrail Details" box when enabled; the same collapsible view is available on the Conversations page. In both places, a "violated" tag is shown in red when a violation is detected, and the box is hidden entirely if the agent has no guardrail configured.

  • Showing Fail Invoke Responses on Testchat & Conversations Page:
    Failed AI Agent invoke responses (any non-200 status) can now be inspected in Test Chat and the Conversations page through a dedicated info box showing the status code, response time in ms, and an expandable response body. The feature is controlled by a new "Show AI Agent Invoke Failure" toggle under Monitoring in Test Chat, and the Conversations page's Show & Hide Details control has been replaced with a gear icon that reveals the same Monitoring toggles, driving visibility on that page accordingly.

  • Pass Delivered Content to Invoke API on User Barge-in:
    When a user interrupts a Voice Agent's response mid-sentence, the next invoke call now includes a barge_in flag along with the exact content that was actually delivered to the user, in original order and unmodified. This ensures the conversation history reflects only what was truly heard, preventing the agent from referencing unspoken sentences and keeping dialogue natural after an interruption. When no interruption occurs, invoke behavior remains unchanged.

Improvements

AI Agents

  • Making Previous Buttons on WebChat Clickable with Configuration:
    A new "Disable previous buttons after a new message" option is now available under the Generic tab in WebChat settings (and as a toggle under a new "UI Interaction" section in Test Chat settings), letting designers control whether previously shown buttons and quick replies remain clickable once a new message arrives. When enabled, buttons preceding the customer's most recent message — including card-based quick replies — become disabled across WebChat, mobile, and Test Chat, preventing repeat selections. The setting defaults to off, preserving current behavior, and quick reply buttons in the Conversations detail screen are always shown as disabled regardless of channel.

  • Navigating to Traces when Clicking a Message:
    Messages in the Conversations page are now clickable, taking users directly to the corresponding trace for that message.

  • Update Subagent Transfer Instructions via MCP Server:
    The Optimization Agent's MCP server now includes a tool for updating sub-agent transfer instructions, and the agent configuration data sent to the Optimization Agent now includes existing transfer instructions. This allows sub-agent routing logic to be refined and kept up to date as part of the optimization workflow, alongside other agent settings.

  • Add ConversationId to AI Testing Conversations:
    Each conversation row in the Evaluated Conversations chart on the AI Testing Evaluations page now includes a copy icon next to its number, letting consultants quickly copy that conversation's ID for further investigation.

  • Async HTTP Request Correlation:
    HTTP Node details in Test Chat and the Conversations page now visually link a request to its matching response: expanding an HTTP Request box highlights the corresponding Response box, using executionId to establish the correlation. This applies to both sync and async requests, making it easier to trace which response belongs to which request during troubleshooting.

  • Adding Prometheus Metrics for Resolution Type:
    A new Prometheus counter metric, knovvu_va_conversation_resolution_total, is now published each time a conversation ends, broken down by tenantId, projectId, channelId, and resolution_type (SelfService, LiveAgentHandover, or NoCustomerInteraction). This allows platform operators to track self-service rates, live agent handover volume, and no-interaction session trends at the tenant, project, and channel level, and to build alerts on top of these metrics.

  • 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.