The May 2026 edition is a big one for builders and for the conversational experience. Headlining this month: Evaluations (Beta) brings structured AOP testing directly into AI Colleagues Studio, and Leena AI can now run as an MCP Server — letting you plug your AI Colleague into external assistants like Microsoft Copilot, IBM watsonx Orchestrate, SAP Joule, or your own custom agent. Real-time response streaming is now live across Web, Slack, and Voice, AI Workflow Generation gets a major usability upgrade, and Onboarding adds no-code Reports. Rounding out the release: native time-based notification audiences (birthdays, work anniversaries, new joiners), a Unified Telephony Channel for SMS and Voice, fuller action support for API skills, in-conversation attachments, a Dashboard Users revamp, accessibility improvements, and a broad set of channel, security, and stability fixes.

In This Edition

---

AI Colleagues Studio

A builder-focused month — from structured AOP testing and a brand-new way to expose your AI Colleague to other applications, to smarter scheduling, fuller API support, and richer attachment handling.

Evaluations (Beta) — Test Your AOPs Before You Ship

AOP testing is now built directly into Studio. You can create test cases, run them repeatedly under realistic user contexts, and track pass/fail results across AOP versions — so you can catch regressions before they reach employees.

AI-assisted test case generation — Generate multiple test cases from an AOP, optionally guiding the AI with free-text instructions and supporting files (policy documents, sample data) so cases reflect real enterprise edge cases. Any individual case can be regenerated with revised instructions if the first pass misses.

Needs Input flow — When a test case depends on documents or data that haven't been supplied yet, it lands in a "Needs input" state instead of failing silently. Reviewers receive an email with a direct link and a guided questionnaire to upload what's required.

Targeted or full-suite runs — Run a selected subset of published test cases or the entire published set.

Multiple iterations per case — Configure how many times each case runs in a single execution to surface flaky or inconsistent behaviour.

Run on behalf of — Choose the context the AOP runs under: system (no user context), a mocked synthetic user (permission-agnostic), or a specific real employee (preserving their permissions and personalised data).

Reporting and drill-downs — Every run captures pass rate, duration, initiator, and the AOP version tested, with history preserved for trend analysis. Drill into any case to see aggregate pass rates, the worst-performing criteria, and plain-language explanations of why something failed — and compare runs side by side to pinpoint where a regression crept in.

Built-in evaluation criteria — Each case is scored against standard quality checks like hallucination, whether all items were processed, correct step order, correct tool calls and parameters, clean user communication, and whether the expected outcome was achieved — with additional workflow-specific and safety checks applied where relevant.

📖 Documentation: Evaluations

Leena AI as an MCP Server

Your AI Colleague can now be exposed as an MCP server, so any MCP-compatible application — Microsoft Copilot, IBM watsonx Orchestrate, SAP Joule, or your own custom agent — can ask it questions, trigger workflows, and run AOPs in plain natural language. This complements the existing ability to connect external MCP servers into Leena as skills, so Leena now works in both directions.

A single conversational sub-agent — Rather than exposing a flat list of tools, Leena presents itself as one conversational agent. The orchestrator picks the right skill, AOP, and knowledge source on its own, so your integration stays stable as the bot evolves.

One-click enablement — Turn on the MCP server from Orchestrator Settings and an OAuth app is provisioned automatically.

Secure, per-user access — Every connection signs in as a real user, so permissions, audience rules, guardrails, and audit logging all carry over.

Credentials managed in-dashboard — Client ID, a one-time client secret with expiry, and redirect-URI management are all handled from the settings page.

Same brain, new surface — Requests run through the same orchestration engine as Web, Slack, and Teams, so capabilities match across every channel.

📖 Documentation: Leena AI MCP Server

AOP Workbench — Scheduling, Lifecycle & Execution Safety

Minute-level scheduling — AOPs can now be scheduled at minute-level granularity, down to a minimum interval of 15 minutes.

New "Completed" status — Workbenches now have a third state — Completed — alongside Active and Inactive. A non-recurring workbench moves to Completed once its last schedule has run, giving admins a clear signal that it's genuinely done rather than just paused or deactivated.

Duplicate execution prevention — If a scheduled run is triggered while a previous run is still in progress, it's safely skipped (with a clear audit reason) so the same items are never processed twice.

API Skills — Full Range of Actions

API-type skills now support the complete set of actions — create, read, update, and delete. Creating, editing, and testing skills all support the new update and delete operations in addition to the existing read and create, covering a much wider range of integration scenarios.

In-Conversation Attachments & Execution Visibility

Attachments in conversation — Employees can now upload files (PDFs, images, and more) directly within a conversation on every channel except phone calls. Both AOP execution and form-filling can read and act on those attachments.

Attachments in execution details — File uploads now appear in the run timeline under both the user message and the AI response — with name, type, and size, plus click-through to preview or download — closing a long-standing gap for runs that involve file processing.

---

Workflows Studio

AI Workflow Generation — Faster, Clearer, Less Guesswork

Building workflows with AI just got a significant usability upgrade. Following its introduction last month, AI Workflow Generation is now faster, more transparent, and more predictable.

Redesigned home page — A cleaner layout that makes it much easier to get started.

Side-by-side visibility — Watch the workflow take shape in real time as you chat with the AI, with no switching back and forth.

A planning step before generation — The AI walks you through its plan before it starts building, so you can course-correct early instead of regenerating from scratch.

Custom naming up front — Name your workflow during planning instead of getting an auto-generated name.

Clearer starting point — The "Manual trigger" is now called the "Start Node," so it's immediately obvious where a workflow begins.

📖 Documentation: Generate Workflows with AI

---

Onboarding

Reports in Onboarding

You can now generate candidate data exports from Onboarding settings — as scheduled emails or as downloadable dashboard reports.

Standard reports, no code required — Previously, reports could only be built as custom reports via API or function. You can now create a standard report by simply picking the columns you want and scoping it with candidate conditions — no scripting required.

Scheduled delivery — Set a schedule and timezone and have the report emailed to your chosen recipients (To/Cc/Bcc), with an option to send a single combined report.

Dashboard filters — Add dropdown, date-range, form-status, action-status, and onboarding-status filters so users can refine the data before downloading.

Custom reports still fully supported — Continue to pull data from forms, tasks, and actions and transform it with a function.

Coming soon: Reports for Offboarding.

📖 Documentation: Reports

---

Platform & Global Settings

A broad set of updates spanning the conversational experience, voice and telephony, notifications, channels, accessibility, security, and overall stability.

Conversational Experience

Real-Time Response Streaming — Following its beta last month, real-time response streaming is now live across Web, Slack, and Voice. Responses now appear progressively as they're generated, rather than all at once after a wait, with new thinking-state animations and streaming-aware controls (quick replies and feedback pause mid-stream), fully localised across 36 languages. On Voice, streamed output is sent straight to the speech engine, cutting several seconds of response latency.

Home as Chat — The new "Home as Chat" experience has shipped, presenting responses on the home screen the same way they appear in chat, with the legacy home experience gracefully retired.

Multiple Tables in Slack — Slack responses now render every table in a response, not just the first.

Voice & Telephony

Unified Telephony Channel — SMS and Voice are now unified under a single channel integration, simplifying setup and management.

Attachment Upload During Calls — The message composer used to be blocked during a voice call, which prevented completing flows that need a file as input. Uploads are now allowed mid-call.

Authentication Beyond Phone-Call OTP — The voice/phone channel now supports authentication methods beyond phone-call-based OTP, so clients with SSO-only policies can deploy voice without workarounds.

Twilio Number Import — Phone numbers can now be imported directly from Twilio, speeding up voice setup for clients already using it.

Richer Voice Instructions — Voice orchestrator instructions now support formatted (markdown) text.

📖 Documentation: Phone Call

Notifications & Audiences

Time-Based Audiences — Admins can now build time-based audiences natively — including "joined in the last N days" (1–365), plus ready-made "Birthday Today" and "Work Anniversary Today" presets — and reuse them across all notification types (bot/push, email, WhatsApp, SMS) from the unified dashboard. Presets evaluate dynamically on the day they run, respect each employee's timezone (with an admin-timezone fallback), and handle edge cases like Feb 29 birthdays and rehire dates. No more manual lists or external tools for recurring birthday, anniversary, and new-joiner messages.

WhatsApp Notification Log Tags — WhatsApp notification logs now carry a taggable identifier (like the one already in email logs), so teams can segregate and identify logs by source.

WhatsApp Template Engagement Metrics — Clicks on interactive WhatsApp template buttons are now tracked correctly (previously they stayed at zero).

📖 Documentation: Time-Based Audiences

Channels & Authentication

WhatsApp OAuth & Approvals — WhatsApp now supports OAuth/SSO as an authentication method, with sign-in links sent over WhatsApp working end to end. Workflow approval notifications are also now reliably delivered on WhatsApp, unblocking approval-dependent processes.

Microsoft Teams Setup & Copy — Removed old, broken Teams setup flows for a cleaner configuration experience, and fixed the "Copy" button so it now reliably copies bot responses — including messages that contain links or emails.

Slack Link Fix — Fixed an issue where links in Slack bot responses could redirect to the wrong page.

📖 Documentation: Microsoft Teams Channel Setup

Dashboard & Accessibility

Dashboard Users Revamp — The Dashboard Users interface has been redesigned for a cleaner management experience.

Accessibility Improvements — Addressed a range of issues raised in customer accessibility audits of the virtual assistant — including accessible names for icon buttons, distinguishable font-size options, visible focus indicators, full keyboard navigation, and correct screen-reader labelling — bringing the chatbot in line with WCAG 2.1 Level A and AA requirements, which is critical for enterprises with accessibility mandates.

Security, Reliability & Bug Fixes

This release also includes a broad round of fixes and hardening:

• Voice — improved call connectivity, transcript ordering, and recovery from network drops.
• Authentication & sessions — more reliable OAuth, token expiry, re-authentication, and login handling.
• WhatsApp & notifications — fixes across template sync, approval-node delivery, audience logic, SMS delivery, and scheduling.
• Slack — better context handling, document recall, and session reset.
• Desktop & mobile apps — more reliable installation and logout, plus a fix for an intermittent "bot agent not found" error when switching bots after an app restart.
• Web — corrected an issue where chats weren't auto-deleting at the admin-configured interval.
• Platform — multiple reliability fixes, a wide query-performance and indexing optimisation sweep, and security and vulnerability remediation across core services.