The April 2026 edition is one of our broadest yet — headlined by real-time Streaming & Thinking State for AI Colleagues, a new Evaluation (Beta) suite that validates AI Colleagues before they go live, a new AI-powered Workflow Builder (Beta) that turns plain-English prompts into working workflows, and Pipedream App which unlocks 3,000+ integrations through a single connector. Knowledge Management gets a major lift with Managerial View in Policy Hub, AI Search, and a native Workday content sync. Helpdesk picks up download-aware filtering, in-line ticket ratings, and a redesigned email-forwarding experience. New first-party connectors land for ManageEngine ServiceDesk Plus and Oracle Fusion HCM, default failure alerts go live for every bot, and the platform sees meaningful gains across chat, channels, voice, auth, and dashboard usability.

In This Edition

Module	Highlights
AI Colleagues Studio	Streaming & Thinking State, Evaluation (Beta)
Knowledge Management	Managerial View in Policy Hub, Workday Content & Permission Sync, AI Search in Policy Hub, Smart Testing Improvements
Workflows Studio	Generate Workflows with AI (Beta)
Helpdesk	Quick Filters on Report Download, Ticket Rating from List View, Improved Email Forwarding, Forwarded Email Replies as Comments
Integrations & Connectors	Pipedream App (3,000+ integrations), ManageEngine ServiceDesk Plus, Oracle Fusion HCM, Default Integration Alerts
Platform & Global Settings	Chat Thread Auto-Renaming, Excel/CSV Upload in Chat, Channel Improvements, Auth & Login Stability, Voice Overhaul, Dashboard Fixes

AI Colleagues Studio

Streaming & Thinking State

The AI Colleague no longer leaves users staring at a "Thinking…" loader. Reasoning, tool calls, source lookups, and the final response now stream into the chat token by token, with a fully expandable thinking timeline that shows exactly how the answer was built.

Faster perceived response — Activity is visible from the first moment the assistant starts working; no more dead-air pauses on longer queries.

Transparent reasoning — Each step of the assistant's thought process — including which skills it invoked and which sources it consulted — is surfaced in real time, building trust and making it easier to audit answers.

PII-safe by design — Sensitive values stay masked while reasoning streams and are only resolved once guardrails approve, so transparency never comes at the cost of data protection.

Resilient to network blips — Transient connection drops are reconnected with backoff and the stream resumes from the last event — users don't have to retry their query.

Currently available in: Web, Desktop App, Mobile App, MS Teams Tab App, and Slack — for tenants on the Chats interface, enabled per bot via configuration. SMS, Google Chat, WhatsApp, and Workvivo continue on their native delivery model.

📖 Documentation: Streaming & Thinking State

Evaluation (Beta)

A new evaluation suite inside AI Colleagues Studio lets AOP authors validate behaviour before a single user is impacted. Test cases are generated from the AOP's own configuration, replayed deterministically with mocked tool outputs, and scored by an LLM-judge on a fixed set of criteria — so failures, regressions, and compliance gaps are caught in pre-prod rather than on the help desk.

AI-generated test cases — Point the generator at an AOP and it reads the prompt, its linked skills, and existing test cases to produce Happy Path, Edge Case, Negative, and Adversarial scenarios. Each test case captures the initiation message, the expected flow, and a saved reference trajectory that every future run is compared against.

Editable, deterministic trajectories — Every tool call in the trajectory exposes its mocked JSON output for editing, letting authors shape exactly what data the AOP "sees" during evaluation. Runs are reproducible by construction — the same test produces the same conditions every time.

Six platform-fixed evaluation metrics — Every run is scored on Hallucination, All Items Processed, Steps in Right Order, Right Tool Calls / Right Parameters, Clean User Communications, and Expected Outcome Achieved. An execution passes only when every metric passes — a deliberately strict bar that catches both hard failures and subtle drift. Metric definitions are platform-managed, so quality criteria stay consistent across teams and AOPs.

Multi-iteration runs and identity context — Any published test case can run up to 10 times per evaluation to surface non-deterministic behaviour. Runs can execute as System or on behalf of a specific employee, so AOPs whose behaviour depends on the initiator's role, location, or data scope can be tested under realistic identities.

Drill-down reports with AI summaries — Run-level, test-case-level, and execution-level views — each with an LLM-generated summary, worst-performing-criteria flags, and per-metric reasoning. Exportable to .xlsx at every level for audit and stakeholder review.

📖 Documentation: Evaluation (Beta)

Knowledge Management

A major lift across Policy Hub, content ingestion, and quality measurement.

Managerial View in Policy Hub

Managers can now see policies assigned to their direct reportees and the answers their team is getting from those policies — turning Policy Hub into a real governance surface. Compliance owners and people managers no longer need to chase screenshots or rely on second-hand summaries to confirm that the right people are seeing the right guidance.

📖 Documentation: Policy Hub Managerial View

Workday Content & Permission Sync

Knowledge Management now connects natively to Workday. Both content and permissions sync automatically — meaning the right people see the right Workday-sourced documents, without manual provisioning or duplicated access rules. Updates in Workday flow through to the AI Colleague's knowledge base on schedule.

📖 Documentation: Workday Connector for KM

AI Search in Policy Hub

Policy Hub upgrades from keyword matching to semantic, AI-driven search. Employees get instant, AI-generated Flash Cards with cited sources — answers in seconds instead of scrolling through PDFs. The pipeline understands intent rather than just literal terms, so phrased queries like "how much paternity leave do I get?" return the right answer even when the policy uses different wording.

Smart Testing Improvements

Knowledge testing now reports against three precise metrics — Completeness, Faithfulness, and Relevance — making it possible to pinpoint whether the AI is hallucinating, omitting key information, or simply being too verbose. Multilingual content handling has also been improved, so testing can be run reliably across non-English knowledge bases.

Workflows Studio

Generate Workflows with AI (Beta)

Workflows Studio now has an AI-powered homepage. Builders describe what they want in plain English — "approval workflow to create a Jira incident from a Slack alert", for instance — and AI assembles the full draft application: forms, nodes, connections, and logic.

AI-powered homepage — The new default landing surface in Workflows Studio. A single prompt produces a complete draft workflow app.

Smart form generation — Mention an integration like "raise a ticket in ServiceNow" and AI reads the connector's configuration to pre-populate the right form fields automatically — no manual field-by-field setup.

AI-assisted action node setup — AI walks through the integration end-to-end: picks the app, asks for the connection, selects the right action, maps fields from workflow context, and auto-tests each action node to confirm it works.

Chat-based refinement — Continue the conversation after initial generation to add nodes, tweak forms, adjust approval logic, or update conditions.

Persistent chat history — Every session is saved. Builders can return any time to resume where they left off.

Good to know: Currently supports form-based (custom) apps; Utility and List View apps remain manual. For Loop, While Loop, and Async Callback nodes are not yet AI-generatable and need to be added in the Flow Builder. Once a workflow is published or manually edited, it's permanently detached from the AI thread and can only be modified through the Flow Builder going forward.

📖 Documentation: Generate Workflows with AI

Helpdesk

Quick Filters Applied on Report Download

Dashboard reports now download with the selected quick filters applied. Exported data matches exactly what's on screen — no more reconciling between a filtered view and an unfiltered export.

Ticket Rating from List View

Users can now rate tickets directly from the ticket list page. Previously, clicking the star ratings redirected users to the ticket detail page; now ratings can be submitted instantly from the list, removing a step from a high-frequency action.

Improved Email Forwarding Experience

Driven by direct user research, the email-forwarding interface has been redesigned. The blocking modal is gone — forwarding now opens within the same scrollable page, so agents can reference ticket details while composing the forwarded message.

Replies to Forwarded Emails Captured as Comments

Replies to forwarded emails are now automatically logged as comments on the originating ticket — even when full email-ticketing is not enabled. The complete external thread stays visible to anyone working the ticket, without manual copy-paste.

Integrations & Connectors

A heavy month for the integrations layer — including a single connector that opens the door to thousands of additional apps.

Pipedream App — 3,000+ Integrations Through One Connector

The Pipedream connector is now live, giving access to a large catalogue of SaaS tools and their actions through a single integration. Instead of waiting for a dedicated first-party connector for every long-tail application, admins can pick any supported app from the Pipedream catalogue and expose its actions as configurable workflow nodes — with input fields sourced dynamically from the underlying API.

Managed authentication, both modes — OAuth, API keys, and token refresh are handled end-to-end by Pipedream. Choose admin-wide auth for system accounts, or per-end-user auth so each action runs in the employee's own context.

Security model — Customers bring their own Pipedream account; Leena AI does not pool customers under a shared account, and the underlying application credentials live with Pipedream rather than on Leena's side.

Use it for the long tail — Pipedream complements the first-party connectors (Workday, ServiceNow, SharePoint, etc.). Use native connectors for flagship integrations and Pipedream to fill the gaps without engineering work.

📖 Documentation: Pipedream

ManageEngine ServiceDesk Plus

A new first-party connector for ManageEngine ServiceDesk Plus is now available, supporting both cloud (OAuth 2.0) and on-premise (API key / auth token) deployments. AI Colleagues can manage IT service desk operations end-to-end — create, update, view, close, and assign requests; add notes and attachments; manage problems and changes; and handle multi-level approvals (add approval levels, add approvers, approve/reject). User-level OAuth is supported, enabling actions scoped to the logged-in technician.

📖 Documentation: ManageEngine ServiceDesk Plus

Oracle Fusion HCM

A new first-party connector for Oracle Fusion HCM is now available, enabling AI Colleagues to interact with Oracle's cloud HCM platform. Supported operations include absence management (create, update, delete absences, check balances, fetch applicable leave types), worker management (hire, view, update, terminate workers), benefits enrollment (view enrollments, check enrollment opportunities), and workflow task actions (approve, reject, withdraw pending tasks).

📖 Documentation: Oracle Fusion HCM

Default Integration Alerts

Pre-defined alert rules are now automatically available for every bot. These default alerts monitor integration and connector execution logs for the most common failure patterns — no manual rule setup needed. Any user with access to Settings → Integrations → Alerts can add email IDs to receive notifications. Alerts route to a fail-safe email address by default and can be redirected to specific recipients.

📖 Documentation: Alerts

Platform & Global Settings

A broad set of platform updates spanning chat, channels, authentication, voice, and dashboard usability.

Chat & Web Experience

Chat Thread Auto-Renaming — Chat threads now rename themselves based on the user's actual query (e.g., "Hiring a Management Intern") instead of staying as generic titles like "Initial Greeting and Help…". Easier to scan, easier to navigate back to.

Excel & CSV Upload in Chat — Users can now upload Excel and CSV files directly from the chat attachment button, opening up AOPs that process spreadsheet data inline.

Channels

Google Chat — Proactive Notifications — Proactive notification delivery is now supported on Google Chat for both individual and admin installations, enabling push campaigns to land where employees already work.

WhatsApp (Infobip) — Dashboard Setup & Notifications — Channel setup gaps are resolved (webhook configuration, phone number validation, template sync). Admins can now create template-based WhatsApp notifications directly from the dashboard, including template sync, variable mapping, and delivery analytics — no engineering ticket required.

📖 Documentation: WhatsApp Setup

Authentication & Login

OAuth Session Stability — Resolved a long-standing issue where the app could fail to locate the OAuth refresh token, causing sessions to silently expire mid-conversation.

Domain-Level Access Control — Closed a bypass where users with personal email domains (Gmail, Yahoo, etc.) could create channels despite strict domain whitelisting. A new flag — Force bot-level authentication for free email domains — routes blue-collar workforces using personal emails through SSO instead of OTP.

Mobile Login Improvements — Fixed Android OTP auto-submit not triggering after auto-fill, resend OTP not auto-populating, a header overlap on the mobile SSO screen, and a client-reported login failure.

Voice

Voice Settings Overhaul — The "Enable Voice" toggle now lives on a dedicated Voice Orchestrator settings page, so voice configuration stays hidden until explicitly enabled. Default selections for Denoising mode and Vocabulary specialization are now correctly applied, and a save-time error has been fixed. New Retell agents now default to Custom mode / Deepgram / 1s endpointing.

Call State Identifiers & Hold Triggers — Voice calls now expose five distinct, real-time states with visual indicators: Connecting, Listening, Processing, Speaking, On Hold. Calls auto-hold when the user opens a webview during an active call.

Voice Instructions Limit — Voice instructions are now capped at 1,000 words to prevent misconfiguration and keep voice agents within the model's effective context window.

Dashboard

Audience Management Fixes — Resolved errors when re-creating an audience with the same name after deletion, missing delete-confirmation snackbars, the "+Audience" button not disabling when no audiences remain, and an incorrect discard popup on duplicate clicks.

User Search by Name — Audience-section search now works with user names in addition to email addresses.

Notification Report Download — The "Download Report" button now appears correctly on the notification analytics page when a notification fails for a single-channel audience.

Changelog

Product Newsletter — April 2026

Product Newsletter — March 2026

Product Newsletter — February 2026

Product Newsletter — January 2026