The
gpt-4.1-nano model for the OpenAI provider has been removed because OpenAI no longer supports it. Migrate to a supported model.2026.23.0
Cognigy
Released November 11, 2026Features
Models Included with Your License
Introduced included models: LLMs and embedding models that NiCE Cognigy provides with your license. You can use them without creating a model or entering API credentials, and their usage is included in the conversation price. Included models are available only for installations integrated with NiCE CXone. Contact your customer success manager to enable them for your organization. Models available in your region are on by default, and admins can activate the others in the admin center under Resources > LLMs.Google Gemini 3.8 Flash Support
Added support for thegemini-3.8-flash model. This model is well suited to agentic workloads that rely on tool calls, structured JSON output, and long-running, multi-step tasks, and its large context window lets it handle long inputs.
Improvements
- Returned the NiCE CXone tenant ID (
cxoneTenantId) and business unit ID (cxoneBusinessUnitId) fromGET /new/v2.0/users/mefor organizations linked to NiCE CXone - Replaced the free-text TLS Option field with a dropdown (
TLS,STARTTLS,None) in SMTP and Email OAuth2 connections. Now, you can see the valid options and select one, instead of typing a hidden value and risking a misconfigured email connection - Added previous and new values to Audit Events in the Admin Center. When you update a Playbook step assertion, Lexicon keyphrase, Knowledge Source, Project setting, LLM, or LLM configuration, the event JSON now includes a
changesarray with the previous and new value of each changed parameter. Credential-like values are omitted - Improved account security by locking user accounts after 5 consecutive failed login attempts by default and emailing users a password reset link to unlock their accounts
- Added OAuth2 authentication for OpenAI-compatible LLM providers. When creating a connection, you can now choose OAuth2 to connect to providers that use short-lived access tokens instead of static API keys
Bug Fixes
- Fixed the issue where deleted or unassigned global LLMs still appeared in Project settings and remained in use
- Fixed the issue where calling
api.say()orapi.output()in a Code Node with a number, for example,api.say(0), sent no visible message - Fixed the issue where using Add Localization on a Node created a localization with the original, untranslated text when the translation service returned an error. Now, an error message appears and no localization is created, so you can try again once the translation connection works
- Fixed the issue where a Say or Question Node wasnโt saved when an adaptive card output with email, URL, telephone, or number in the
stylefield contained CognigyScript - Fixed the issue where Microsoft Azure OpenAI content-filtered requests with streaming on the Responses API were returned as a normal answer instead of a content filter error
- Fixed the issue where the Retrieve Knowledge Node returned unfiltered results in chunks mode when filters contained CognigyScript expressions or invalid JSON
- Fixed the issue where importing a Package failed with
Invalid value for field 'fields'when the Package contained a Connection whose credential had been revoked - Fixed the issue where services kept memory for every RPC call that timed out, which made them run out of memory over time
- Fixed the issue where
mistral-medium-latestwas incorrectly labeled as deprecated in the model picker, despite still being supported by Mistral - Fixed the issue where users with only the user manager role saw a blank Access Control page in the admin center and an empty Assigned Projects section on the user details page
- Fixed the issue where adding a user to a Project they were already a member of caused their Project roles to be ignored, and where adding a Project member with a Management API key returned a
404error although the member was added - Fixed the issue where new SSO users were created with only the base role when the identity provider registered more than one role. Now, all valid roles are assigned
- Fixed the issue where a user deleted in NiCE CXone was created in Cognigy as an enabled user when Cognigy didnโt know the user yet
- Fixed the issue where the Flow list kept showing the previously selected Projectโs Flows after switching Projects
- Fixed the issue where a Flow reached through an Execute Flow or Go To Node on a Snapshot failed with
No LLM Provider configured, even when an LLM was correctly assigned to the Project - Fixed the issue where AI-generated simulation personas started with stray labels such as
role:or"role":"user"
Cognigy Voice Gateway
Released November 11, 2026Improvements
- Improved service stability and performance
Cognigy Webchat
Released November 11, 2026Improvements
- Added user inactivity detection to Webchat v3: new Enable User Inactivity and Disable User Inactivity Flow Nodes and an endpoint User Inactivity setting let you detect a silent user and, after a configurable timeout, inject configurable text and data input into the current Flow or switch the conversation to a selected Flow, so you can re-prompt or end abandoned conversations. Detection and injection are reported in Cognigy Logs
Bug Fixes
- Fixed the issue where minimizing the Demo Webchat in the centered layout hid it until the page was reloaded
Cognigy Agent Copilot
Released November 11, 2026Bug Fixes
- Fixed the issue where the first Agent Copilot Whisper message appeared as a user-facing chat message after a handover
Cognigy Live Agent
Released November 11, 2026Bug Fixes
- Fixed audio and video playback from external URLs in conversations
Cognigy Insights
Released November 11, 2026Bug Fixes
- Fixed the issue where analytics records were not saved when an intent exceeded 255 characters, causing other records saved in the same batch to be lost
- Fixed the issue where sessions transferred with the Voice Gateway Transfer Node (
SIP REFER) were counted as 0 handovers. As a result, Insights reports and OData Sessions showed fewer handovers than actually occurred - Fixed the issue where Insights showed incorrect Flow execution times. For example, a Flow that took 3 seconds was recorded as taking only a fraction of a second
Infrastructure
Cognigy + Insights + xApps + Agent Copilot
On-premises Upgrade Requirements for v2026.23.0
Before upgrading to Cognigy 2026.23.0, review the following requirements.-
MongoDB Helm Chart version. The images used by Cognigy MongoDB Helm Chart 15.6.29 are hosted in
cognigybitnami.azurecr.io, which on-premises installations canโt access. Remain on MongoDB Helm Chart 15.6.28 to avoid image pull failures. Cognigy 2026.23.0 is compatible with both chart versions. -
db-init image version. The new default
db-initimage is hosted incognigybitnami.azurecr.io, which on-premises installations canโt access. If thedb-initJob cannot pull its image, the Helm upgrade will fail. Before upgrading to v2026.23.0, override thedb-initimage in yourvalues.yaml:If you use a custom MongoDB image, for example, one hosted in your own registry, configure the following options:imageโ the full image reference. The image must includebashandmongosh, or the MongoDB Atlas CLI if you use MongoDB Atlas (mongodb.scheme: mongodb+srv).registryโ the registry to use for this image. It overrides the registry specified inimageand takes precedence overglobal.imageRegistry. Set this option ifglobal.imageRegistrypoints to a registry that does not contain the required image.imagePullSecretsโ the image pull secrets used by thedb-initjob. Specify the Secret containing your registry credentials. The Secret must exist in the Cognigy namespace.