The July 2026 edition is a heavy one. AI Colleagues Studio gained Webhook Triggers, so business processes can now be started by external systems rather than only from chat, alongside Studio Assist (BETA) — an AI debugging partner built into the Studio — and real-time voice streaming that cuts the silence before the AI Colleague starts speaking. Knowledge Management picked up the ability to stop a sync mid-run, plus a full layer of failure visibility through connector execution logs, wider failure alerts, and a zero-article safety net. Workflows Studio can now build Employee Sync apps with AI on a redesigned canvas, IT Automation added editable install scripts for macOS and PowerShell support on Windows, and Notifications gained archiving along with a "send overdue instantly" option for timezone-aware delivery. Platform updates span channel improvements, security and access hardening, web widget uploads, mobile, and a broad set of stability fixes.

In This Edition

---

AI Colleagues Studio

A significant release for the Studio — AOPs can now be triggered by external systems, a new AI assistant helps you diagnose runs without digging through them manually, and voice conversations feel markedly faster.

Webhook Triggers (v1) — Start AOPs from External Systems

Until now, an AOP ran when someone asked for it in chat or when a schedule fired. A new Triggers tab in Workbench adds a third route: an inbound webhook, so an AOP can start the moment something happens in another system.

Two run modes — AOP run conditions now cover On a schedule or via an External trigger.

Simple setup — Webhook configuration gives you a Production URL to paste into your external system, along with a copy-ready request example. Only POST is accepted.

Authentication that you control — Access is enforced through scoped API credentials. Only active credentials carrying AOP-execute access can be attached to a webhook, so revoking a credential immediately cuts off the systems using it.

Test before you trust — A dedicated Test URL never initiates an AOP. Click Start listening, send a sample event, and verify your setup live. Captured events show exactly what was received; rejected ones show precisely why — invalid authentication, malformed payload, rate limit, wrong method, or an oversized request.

Visible in Run History — Webhook-triggered runs appear alongside every other run, with a new Trigger type filter (Manual · Schedules · Webhook).

App Event triggers — running an AOP when a matching event is detected in a connected app — are coming soon.

📖 Documentation: Setting Up Triggers

API Credentials — A New Settings Page

Admins can now create scoped credentials for external systems that call Leena AI APIs, and revoke them to cut off access immediately. Credentials are the backbone of webhook authentication above, and are managed in one place.

📖 Documentation: API Credentials

Studio Assist (BETA) — Your AI Debugging Partner

Working out why a run behaved a certain way used to mean combing through run history, applying filters, and reading step payloads one by one. Studio Assist replaces that with a conversation.

Ask, don't dig — A new Studio assist panel opens from the top bar across AI Colleague surfaces — the AIC listing, Run History, Run Details, and the Debugging Console — and picks up the context of whichever page you're on. Ask in plain language: "why did this run fail?", "show me unhandled runs for this AOP from yesterday."

Diagnosis with evidence, not guesses — Studio Assist investigates the actual run data — run history, step execution, tool inputs and outputs, and reporting insights — and explains the root cause with the supporting evidence behind it. Answers stream in live and can include structured tables and timelines, not just text.

It takes you to the problem — Responses deep-link straight to the relevant run, step, or Studio page, so you go from question to fix without navigating back yourself.

Product questions too — It also answers how-to and configuration questions grounded in Leena AI documentation, right inside the Studio.

A real conversation — Context-aware starter suggestions on every surface, follow-up prompts after each answer, chat history with resumable sessions, and per-answer feedback.

Safe by design — Studio Assist is read-only. It investigates and recommends but never changes your configuration. PII is masked in what it reads and what it returns, access respects your existing Studio role permissions, and platform guardrails apply to every conversation.

📖 Documentation: Studio Assist

Voice — Faster, More Responsive Conversations

The biggest change to the voice experience this release is that responses now stream in real time. The AI Colleague starts speaking as soon as the answer begins generating rather than waiting for the complete response, which noticeably reduces the pause before you hear anything — most visible on questions that draw on knowledge or kick off a workflow.

No dead air while work happens — When a request needs processing, the acknowledgement is spoken in parallel with the task being dispatched, instead of after it.

Instant recall — Repeat requests, status checks, and follow-ups on already-shared information are answered straight from the conversation, with no backend round-trip.

Graceful interruptions — When you interrupt mid-sentence, the AI Colleague knows exactly what you heard and picks up from there, without repeating itself or dropping information.

Reliable short answers — Single-word replies such as "Yes," "No," and "Wait" are now understood consistently.

Alongside these, a round of polish: the AI disclaimer is no longer read aloud mid-call, Listening and Speaking indicators reflect the correct state, the employee-details form no longer loops, call analytics show accurate data, autonomous agent voice setup no longer errors out, and common IT phrasing such as "my PC is running slow" now routes to the right place.

📖 Documentation: Voice Architecture

---

Knowledge Management

This release focuses squarely on control and visibility over connector syncs — being able to stop one, understand why one failed, and make sure the right people hear about it.

Stop a Running Sync

A sync that's pulling the wrong content or running against the wrong configuration no longer has to be waited out. A new Stop Sync option under the three-dot menu on the Integration Details page cancels a connector sync that is currently in progress.

• Selectable only while a sync is actually running, with a confirmation step before anything is halted
• Cancellation carries across the connector, Knowledge Management, and the parser — remaining batches are rejected and no further data is processed
• Status updates to Sync Cancelled consistently across the dashboard, integration details, status pills, and sync logs
• Applies to connectors on the new sync architecture; REST, SFTP, and Workday are not in scope

📖 Documentation: Stopping a Sync in Progress

Failure Visibility — Know When a Sync Breaks, and Why

Connector execution logs — A new KM Connectors tab under Integrations → Execution logs surfaces the detail behind every sync request. Columns cover connector, connection name, sync ID, timestamp, trigger type (manual or automatic), request method, request URL, status, and status code. Opening a row exposes the metadata, request body, and response body — so the actual API error is visible directly, without needing a backend investigation. Filters are available on sync date range, connector, connection name, trigger type, and request method.

📖 Documentation: Connector Execution Logs

The right people get told — When a connector sync fails, notifications now also reach the assigned account team for the affected assistant, on top of the existing recipients (setup user, owner, and any added recipients). Because the notification is resolved across all of a customer's environments, the same account team is alerted regardless of which environment the failure occurred in.

Zero-article safety net — If a sync returns zero articles, it now fails rather than reporting success, and existing articles are left untouched instead of being removed. This closes a scenario where a misconfigured or broken source could quietly empty a knowledge base. Applies to connectors on the new architecture; REST and SFTP are not in scope.

📖 Documentation: When a Sync Finds No Articles

Credential Security

Secrets entered during connector setup are now masked and cannot be retrieved once saved — the only option is to replace them. This brings credential handling in line with how enterprise platforms such as Okta and Zendesk treat the same data.

Separately, the SharePoint Graph API migration banner now appears only for connections still on the older setup, with corrected messaging and a direct link to the migration guide.

Sync Improvements

• Faster audience-only updates — Changing only an article's audience now completes in under half the time it previously took, with updated and deleted audiences correctly reflected in answers
• ServiceNow KB Number — The ServiceNow connector now fetches the KB Number as metadata, so articles can be searched by KB number
• Additional columns in Connector Sync Logs, plus relabelled and reordered columns on the connectors table
• Configuration for an already-configured connector can now be viewed for any step directly, without moving through the setup screens one at a time

📖 Documentation: ServiceNow Integration

End-User Experience

Download source articles in their original format — Employees can now download a source article as the file it actually is. Non-PDF files such as .docx download as-is, in both Sources and the Policy Hub webview.

Better acknowledgement emails — Employee acknowledgement emails now carry a View Policy action that deep-links to the article in the web app, and the subject line and signature use the assistant's configured title rather than an internal name.

Smaller Improvements & Fixes

• Image handling during parsing is now model-driven — the system decides which images actually matter to an article instead of treating all of them the same
• The Article expiry date helper text no longer states that articles are auto-archived; archiving is governed by the separate Move articles to Archive state after expiry setting
• Smart Testing: the document preview now clears when an applied filter excludes the previously selected article

---

Workflows Studio

Employee Sync via AI

Employee Sync utility apps can now be created — and refined — directly from the AI builder, with no manual wiring. Describe the data source in plain English and the AI classifies it as an Employee Sync utility, confirms how the source paginates, and builds the required scaffold for you.

• Automatically configures the fetch and transform steps along with the Pagination & Search form, including Text, Dropdown, Master, and Dynamic Dropdown fields
• Edit via AI now works on Employee Sync apps too — previously these could only be built and changed by hand

📖 Documentation: Generate Workflows with AI

Redesigned Flow Builder Canvas

The Flow Builder canvas has been redesigned — cleaner, and easier to navigate as apps grow.

Paste as cURL in the REST API Action

Setting up a third-party API no longer means filling in fields one at a time. In an Action node set to REST API, paste in a cURL command and the request auto-fills — method, URL, headers, query parameters, and body. Useful for wiring up an API in seconds straight from its documentation or an API client export.

---

IT Automation

Editable Install Scripts for macOS Software

macOS software now comes with an editable installation script. A default script is generated based on the uploaded file type (.pkg, .dmg, .app, .zip) and can be reviewed and edited before you continue — so packages that need a non-standard install step no longer need a workaround. macOS software created before this change keeps its earlier behaviour and shows no editor.

PowerShell Support for Windows Scripts

Windows scripts now have a Script type field: Batch (the default) or PowerShell. Existing scripts are treated as Batch, so no action is needed on anything already configured. macOS scripts continue to use bash.

Documentation has been updated with Windows and macOS configuration split out — accepted file types, how install scripts are produced per operating system, Mode × Account type behaviour, and OS-specific troubleshooting.

📖 Documentation: Scripts · Documentation: Softwares

---

Notifications

Archive Notifications

Completed, cancelled, aborted, and draft notifications can now be archived from the row menu, clearing them out of the Single Send and Recurring lists without deleting anything.

• A new Archived tab, with Archived by and Archived on columns
• Unarchiving returns a notification to its original tab with its original status
• Archiving a recurring schedule cascades to all of its child runs; unarchiving reverses it
• Active notifications (pending, scheduled, running) can't be archived — cancel them first
• Permissions match Cancel, so anyone who can cancel a notification can archive it

Send Overdue Instantly — Timezone-Aware Delivery

Timezone-aware delivery splits an audience by timezone so each person receives a notification at an appropriate local hour. Until now, anyone whose local send time had already passed (beyond a short tolerance window) was skipped entirely. A new checkbox under Timezone-Aware Delivery controls what happens instead:

• On (default) — those users receive the notification immediately rather than being skipped
• Off — the previous behaviour, where they are skipped

Available on Standard, Email, and WhatsApp notifications, and editable on an existing notification from its detail page. Choosing a filter-based audience turns Timezone-Aware Delivery on automatically, so the setting applies there too.

Note: Notifications created before this shipped don't carry the setting and keep the earlier skip behaviour until they are edited and saved.

Filters & Accuracy Improvements

• Filter the notifications list by Type and Channel
• The audience builder now handles array-typed sync fields
• A sweep of accuracy fixes: success and failure totals are no longer double-counted, recurring notifications no longer flip to "completed" while sub-runs are still going, timezone configuration now persists correctly on save, and the Recurring grid re-sorts by Scheduled on as expected

---

Platform & Global Settings

A broad set of updates across channels, security, the web widget, email, and mobile.

Channels & Chat

Workvivo, done right — Markdown now renders cleanly, and tables and images come through as attachments rather than garbled text.

No more silent waits in Slack — With streaming enabled, users watch the assistant respond in real time instead of waiting on a static screen.

Streaming that survives restrictive networks — The app now detects SSL-inspection proxies that buffer responses and walks users through the fix, rather than leaving them looking at a frozen screen.

Security & Access

SAML SSO sign-in hint forwarding — Sign-in can now be pre-filled from the identity provider, reducing friction on SSO-first deployments.

Screenshot gating on the mobile app — A new per-assistant setting blocks screenshots, enforced natively in the mobile app for organisations that need it.

Per-tenant control over white-labelled login gating — Gating behaviour can now be configured per tenant rather than globally.

Tighter access control on assistant-served documents — Access controls and caching behaviour have been tightened for documents served through the assistant, so sensitive files such as payslips are protected end to end.

SSO reliability — Fixed SAML test-login failures, and token-expiry timeouts now honour the configured interval.

Web Widget & Uploads

Bigger, more uploads — A higher file size limit, and more files at once.

Also fixed: a phantom chat bubble appearing when nothing was processing, an unresponsive Refresh Page button on the microphone modal, embedded links refusing to open in Safari, and a duplicated footer image on mobile.

Email & Messaging

Disclaimers on SMS — An optional disclaimer can now be appended to text messages.

Consistent assistant identity — Your assistant's configured identity now appears correctly in email titles and signatures.

Email log statuses now read in one consistent tense.

Mobile

• White-labelled iOS and Android builds are live for enterprise deployments, fully customer-branded
• Fixed a missing iOS notification sound and a biometric toggle that required two taps

Stability & Reliability

• Timeout and retry handling added to a critical access-control call in core, so a transient blip doesn't cascade
• Slack scheduled-job setup hardened against intermittent errors
• Improved error classification in monitoring, so genuine issues are separated from network and browser noise
• Fixed user-group creation failing silently, employee-sync termination edge cases, and audience syncs getting stuck in a "running" state

Client-Reported Fixes

A batch of account-specific issues cleared across WhatsApp (welcome message loops, template errors, SSO and OAuth), Microsoft Teams (channel users and notification delivery), Slack (assistant authentication for blocked users), and persistent login and session loops affecting several deployments.