Helvia.ai Release 2026.05.07

07 May 2026

1. Authenticate Users Mid-Conversation with OIDC on Webchat

You can now securely authenticate end users during a conversation using OpenID Connect (OIDC), directly within your AI Agent workflows. By adding the new Authenticate User node, you can prompt users to log in via your configured identity provider or validate existing tokens for a seamless experience. If the user is already logged in (e.g. via Microsoft SharePoint), their token is automatically validated, avoiding redundant login steps. Once authenticated, user claims (such as identity or roles) are made available in the workflow, enabling more personalized and secure interactions.

Why it matters: Previously, authentication had to be handled outside the conversation or required custom implementations. With this update, you can enforce secure access to sensitive actions or data exactly when needed, without disrupting the user journey.

Example use case: A customer support AI Agent asks users to authenticate before showing account details or processing requests.

How it works: Configure your AI Agent’s identity provider in Security Settings (OIDC).

Then, insert the Authenticate User node into your workflow.

The platform handles the authentication flow (redirect or token validation) and automatically resumes the conversation once the user is verified, providing authentication data for use in subsequent steps.

2. Global Interaction Logs for End-to-End Debugging

You can now access a centralized Interaction Logs page in the Observatory, giving you a complete, cross-session view of all bot interaction metadata in one place. Instead of navigating individual sessions, you can monitor, filter, and analyze interactions across all agents and deployments from a single, unified table.

Why it matters: Previously, debugging required drilling into individual chat sessions, making it time-consuming to trace issues across conversations. With this global view, you can quickly identify errors, track behavior patterns, and troubleshoot flows more efficiently.

Example use case: An AI Agent designer investigating a failed workflow can filter interactions by agent and date, locate error events instantly, and jump directly to the relevant session to understand what went wrong—reducing debugging time significantly.

How it works: Navigate to Observatory → Interaction Logs. Use filters such as date range, agent, or Session ID to narrow results. Click on any row to view the full interaction payload in a structured JSON viewer, or use the View in Session link to jump directly to the exact step within the conversation.

3. Knowledge Base Enhancements

3.1 Enhanced Knowledge Base Segmentation with AI Controls

You can now fine-tune how your documents are segmented into Knowledge Base articles by advanced SDS (semantic-doc-segmenter) capabilities directly in the Console. This includes AI-powered parsing, article sizing, and image extraction controls, giving you more flexibility and control over ingestion quality.

Why it matters: Previously, document segmentation provided limited control over how content was broken down and processed. With this update, you can optimize ingestion for accuracy, cost, and retrieval quality—choosing between fast standard parsing or advanced AI-driven extraction depending on your use case.

Example use case: A legal team uploads complex PDF contracts and enables AI-Powered (Agentic) mode with image extraction and large article sizing to preserve full contextual meaning. Meanwhile, a support team uses Standard mode for faster, cost-efficient processing of FAQs.

How it works: In the Upload File modal, you can switch between Standard (Fast) and AI-Powered (Agentic) modes. When Agentic mode is selected, you can configure article size (Small → XLarge), enable image extraction (for PDF/DOCX), and optionally add processing instructions. The same configuration can also be applied at the integration level (Azure Blob and SharePoint), ensuring consistent segmentation rules for all synced content. All settings are forward-only and apply to new uploads or syncs.

3.2 Automatic Source Attribution for Knowledge Base Articles

Knowledge Base articles now automatically include source attribution metadata, ensuring every generated segment carries its origin information without manual configuration. This includes the original document URL as well as page-level references where available.

Why it matters: Previously, source tracking had to be configured manually. With this update, every article is automatically traceable to each original source, improving transparency and auditability.

Example use case: A compliance team reviewing AI-generated answers can instantly see which SharePoint document and page range each Knowledge Base article originated from, enabling faster validation during audits or regulatory reviews.

How it works: During document ingestion, the system automatically propagates sourceUrl and page from the document pipeline into each Knowledge Base article. For PDFs, DOCX, and PPTX files, page numbers are included where available. In the Console, this metadata is visible in the article detail view, while sync-based articles maintain read-only source attribution for consistency and traceability.

4. Access Detected User Language in Agent Flows

You can now access the detected user language inside your agent workflows via a new system variable, even when that language is not supported by the agent. The new detectedLanguage variable exposes the result of the language detection plugin independently from the agent’s active language configuration.

Why it matters: With this update, AI agents can explicitly identify unsupported languages and respond appropriately instead of incorrectly continuing in a fallback language.

Example use case: A customer writes in Japanese to an agent configured only for Greek and English. The system detects the language as detectedLanguage = "ja", allowing the agent to respond with a message such as: “Sorry, I can only assist in Greek or English.” This improves clarity and avoids misleading responses in unsupported languages.

How it works: The Language Detection plugin runs during message processing and provides its raw detection result through detectedLanguage. This variable is session-scoped, read-only, and always reflects the most recently detected language, regardless of whether it is supported by the agent configuration. If no detection has occurred or the plugin is not enabled, the variable returns an empty value.

This feature requires the Language Detection plugin to be enabled first.