New Features 🚀​

Proxy Agents as OpenAI-Compatible Endpoints​

Serenity* Star Proxy Agents are now available through OpenAI-compatible Responses and Chat Completions endpoints. Use any model in Serenity* Star through a familiar interface, and plug your agents straight into Copilot Chat or any OpenAI-compatible client with no custom integration work.

Every request keeps full audit tracking inside Serenity* Star and runs through Content Safety, Skills and usage limits, so you get the reach of an open standard with the governance and control of the platform.

Serenity* Apps​

Serenity* Star now supports integrated Apps inside Serenity* AI Chat. Apps are a new way to work with agents: instead of a plain conversation, each agent gets a streamlined, purpose-built interface focused on the task at hand. By putting value and functionality first, Apps make it easier for your organisation to turn agents into everyday productivity tools.

Compliance Document Certification App​

The first App in the catalog is the Compliance Document Certification App. Upload a document, contract, or any internal file and validate it against a defined set of rules.

An AI agent evaluates the document and returns supported corrections grounded in its content and the applicable rules. Iterate on the document, apply corrections, and explore suggested edits until it meets your compliance standards.

---

Enhancements ✨​

Model Update Notifications​

Stay informed when model provider changes affect your agents. If a parameter your agent uses is no longer supported, you'll receive an email, and recommended configurations are shown clearly in the UI so you can keep your agents performing at their best.

Runtime Skill Execution​

Control exactly which skills an agent can use on a per-execution basis. Include or exclude skills at runtime by specifying them in the request, and the agent will use only the ones you allow, giving you precise control over agent behaviour for each call.

Sensitive Data Obfuscation​

Sensitive data obfuscation now offers finer-grained control. Enable obfuscation individually for agent behaviour or for input parameters, giving you more flexibility and ensuring consistent, predictable handling of sensitive data.

Context Improvements​

The context engine has been revised to allocate unused space more intelligently. Smarter budget handling lets agents produce better responses without compromising the context they rely on.

Security​

We remain committed to continuously strengthening security across every layer of the platform. HTTP requests and file uploads via URL are now protected against untrusted sources, further safeguarding your data throughout Serenity* Star.

---

Bug Fixes & Minor Improvements 🛠️​

• Updated the AI Chat default models for skills to more recent and powerful models.
• Fixed an error in the model selector side panel in the Vision service view.
• Improved image visualisation in the Bring Your Own Model functionality.
• Audio generation instances are now shown with the correct service type.
• The service instances grid now shows the model used in each execution.

New Features 🚀​

New AI Services​

Vision Service​

Serenity* Star now includes a dedicated Vision service you can use directly from the API or in the playground. Automate image analysis and get detailed, structured insights without wiring up a full agent. It's a fast way to add image understanding to your workflows and integrate the results into your own applications.

Text to Speech​

A new Text-to-Speech service lets you generate high-quality speech audio files straight from the API or in the playground. Turn any text into natural-sounding audio on demand.

Skills​

Prompt Skill — New Mode​

The Prompt skill now supports a new Instruction Mode. It lets the agent read additional instructions on demand for extra context. This keeps your main behaviour lean while still giving the agent the guidance it needs, exactly when the situation calls for it.

Vision Skill​

The new Vision skill gives any agent the ability to analyze and understand images within its flow. Add vision to models that don't support it natively, or bring in a more capable vision model when the situation calls for it. Your agents can now inspect screenshots, documents, and photos as part of a task and act on what they see.

Chat Widget Reasoning​

The Chat Widget now can show the agent's reasoning, giving end users better insight into how each response was produced. Showing the steps behind an answer builds trust and helps users understand the logic that led to it, right inside the conversation.

Showing reasoning must be enabled in the Chat Widget channel configuration.

---

Enhancements ✨​

Web Search — New OpenAI Models and Citations​

Web Search now supports the latest OpenAI models, bringing more capable and up-to-date reasoning to your agents' searches. Answers that draw on Web Search results are supported for in citations, so users can trace each response back to its sources and verify where the information came from.

Reasoning Tokens in Instances​

The instance detail now reports reasoning tokens, giving you clearer visibility into how reasoning models consume tokens. With this breakdown you can better understand and forecast the cost of reasoning-heavy executions across your agents.

Successor Model Email Notifications​

Get notified by email when a successor model is announced, keeping you up to date with the latest, state-of-the-art models. Plan your migrations ahead of time and adopt newer, better-performing models before older ones are retired.

Volatile Datasets and Excel Support​

Datasets now support Excel files. Compatible CSV and Excel files attached as volatile Knowledge are automatically converted, when their schema is suitable, into a volatile dataset that the agent can query on demand, improving accuracy, context usage, and response quality.

Bring Your Own Embedding Model​

The Bring Your Own Model feature now supports embedding models through OpenAI-compatible endpoints. Connect your own embedding infrastructure to power Knowledge search keeping data within your control, strengthening compliance and governance.

---

Bug Fixes & Minor Improvements 🛠️​

• Updated the JavaScript and .NET SDKs to use the new per-agent Volatile Knowledge endpoints.
• Improved handling of vendor errors.
• Improved logging of skill failures caused by unhandled errors.
• Fixed an incorrect Knowledge separator indicator that was not producing section breaks.
• Fixed an issue with enabling/disabling "enable thinking" on Alibaba models.
• Performed a full review of AI model configurations to ensure all capabilities are up to date, and fixed several vendors to support advanced thinking capabilities.
• The Demo Page side panel now shows a link to copy the page URL.
• Added a streaming indicator in the instance detail.
• Fixed an issue that prevented sorting by price in the Store.
• Enabled Adaptive Thinking configuration for Anthropic models.
• Added the ability to delete your own models when they haven't been used, along with the option to enable/disable them.
• Fixed an issue that prevented loading the same AI model with a different mode for a vendor.
• Adjusted the services catalog to group services by type.
• Improved JSON extraction for the jsonContent field in agent responses.
• Fixed an issue with the fallback model when the same vendor was configured for both models.
• The Serenity* Store now supports Markdown in your listings for clearer, richer presentation.
• API Key Pooling is now available for embedding services, providing more stable operation.
• We continue to strengthen platform security, with enhanced encryption of sensitive data throughout Serenity* Star.
• Serenity* AI Chat now initializes with ready-to-use skills, with no manual configuration needed.

New Features 🚀​

Sensitive Data Obfuscation​

Serenity* Star Agents now support a new Sensitive Data Obfuscation mode. Based on OpenAI's privacy filter and hosted natively in Serenity Cloud, it redacts sensitive information from your LLM requests inside private, secure infrastructure — strengthening compliance and governance without sending data to third parties.

Get Agents from the Serenity* Store​

You can now get agents directly from the Serenity* Store. Browse the catalog, explore resources shared by the community, and purchase or get agents from your favorite publishers — then put them to work with little to no setup. This completes the Store experience previewed in the previous release.

Citations​

Citations are now available across Serenity* AI Chat, the Chat Widget, the Agent Design Studio preview, and Agent Instances. Show users the exact passages from your Knowledge Sources behind each answer to build trust, and use the new citation usage tracking to understand how your agents draw on knowledge. To learn more, refer to the Knowledge documentation.

Serenity Inference Endpoint​

Serenity* Star Orion models are now available through the API as Open AI Compatible Endpoints. Check out the API Reference for more information.

This feature is currently in preview.

ElevenLabs Integration​

Serenity* Star now integrates with ElevenLabs, giving your agents access to high-quality text-to-speech and speech-to-text models directly within the platform.

---

Enhancements ✨​

Email Sender — Dynamic Content​

The Email Sender skill now supports a Dynamic Content mode. Let the agent fill in specific parts of your subject and body while keeping the rest of your template exactly as you intended — combining the control of a static template with the flexibility of AI-generated content.

Security​

Serenity* Star continues to strengthen security at every layer:

• Hardened the Knowledge processing endpoints against denial-of-service (DDoS) attacks.
• Improved security in the HTTP Request skill to block requests to unwanted or unsafe endpoints.

Demo Page Controls​

Gain full control over your Demo Page availability. Deactivate the current page or regenerate its link at any time to prevent misuse.

Connectors​

The new Connectors tab in the Compliance Center gives you a full picture of every integration in use across your organization. Quickly see where users are connecting their Serenity* accounts and which agents rely on each integration.

---

Bug Fixes & Minor Improvements 🛠️​

• Added new locally hosted embedding models, including Serenity Cloud embedding models.
• Fixed an issue running skills with Gemini models.
• Fixed vision support for Anthropic models.
• Improved agentic file processing to use vision when images are present.
• Knowledge file metadata is now copied when importing or duplicating an agent.
• Added a system to strip invisible characters that caused certain Knowledge files to fail processing.
• Fixed validation of JPEG files in volatile Knowledge.
• API Keys are now obfuscated in the Bring Your Own Model configuration and upload form.
• Increased the allowed size for agent system definitions.
• Added a disclaimer in Citations noting that results may vary depending on the model.
• Added empty states to the Marketplace.
• Fixed an issue where the skill title wasn't shown in the instance detail when an execution failed.
• Fixed an overflow issue in the Agent Manager.
• Fixed an execution error in the Skill Emulator.
• Fixed form sizing issues.
• Fixed an error in the deprecated-models notification job.

New Features 🚀​

New Serenity* Star Engine​

We've rebuilt the foundation that runs your Agents. The new Serenity* Star Engine is faster and more stable, keeps pace with the latest advances in agentic systems, and gives us a stronger base to build on going forward. The rebuilt architecture improves how Agents reason over multiple steps, call skills, and stream responses — and it lets us ship new capabilities to you faster. No action is required — every Agent now runs on the new engine automatically, with no changes to your existing configurations or integrations.

Reasoning visualization — An Agent's reasoning flow can now be visualized, giving you better insight into how each response was produced.

Improved skill logging — Skill logs now clearly distinguish between parallel and sequential executions, making it easier to follow what your Agent did and when.

Serenity* Store & Marketplace​

Buy and sell agents across the Serenity* Star ecosystem. Use the Serenity* Marketplace to publish your own agents and earn usage credits on every purchase. Then browse the Serenity* Store to find ready-to-use agents and put them to work with little to no setup.

This functionality is currently in preview. The ability to get agents from the Store is coming soon.

Bring Your Own Model​

Bring your own LLMs into Serenity* Star. Run models hosted on your own infrastructure for greater governance and to meet your compliance requirements.

Bring Your Own Model requires an OpenAI-compatible inference endpoint.

Citations​

Agents now support citations. When a response draws on your Knowledge Sources, it can show exactly which source each part of the answer came from — giving you clearer traceability and greater trust in agent output. To learn more, refer to the Knowledge documentation.

Citation results are currently only available via the API. Support for Serenity* AI Chat, Agent Preview, and the Chat Widget is coming soon.

---

Enhancements ✨​

Guardrails​

RagMetrics Guardrails now let you select the evaluator model. Tenants can configure their own RagMetrics account and choose the evaluator model that best fits their needs.

Input Parameters Hook​

The Chat Widget now provides a getInputParameters hook. Pass a custom function to supply updated input parameters with each message sent. Refer to the Chat Widget (JavaScript) documentation for more details.

Knowledge​

• Knowledge Source metadata can now be generated or updated automatically with the "Generate with AI" button in the metadata section.

• The file types allowed in volatile Knowledge can now be configured per agent, so you control which file types users can attach to a conversation.

---

Bug Fixes & Minor Improvements 🛠️​

• Improved performance of aggregate statistics in the Agent Manager and Cost Dashboard.
• Adjusted parameter validation in the Agent Design Studio when sending a message.
• Improved positioning of select components that could render off-screen.
• Improved empty states in the Agent Design Studio so they take up less space.
• Fixed an error that occurred when changing the Guardrails validation method.
• Dataset validation now runs on the replace and append API endpoints to return a clear error to the user.
• Renamed the RagMetrics "Legal Advice" category to "Compliance Awareness".
• Improved anti-forgery token regeneration in the Agent Designer, preventing an authentication error when the token expired.
• Improved parameter validation in the Agent Design Studio API
• Fix a validation error when changing evaluation method in Guardrails configuration
• Improved schema validation in Dataset API endpoints when doing replace and append
• Improved connection setting pooling in fallback models

New Features 🚀​

Owner Access API Key​

Create a Personal Access API Key. This key inherits all user roles and permissions for enhanced flexibility in the API. Changes in user permissions are auto-updated, so the API Key always works as an extension of the owner. You can now manage integrations and automation with greater security and ease, as the key always reflects your current access rights.

For more information refer to the Developer Tools Documentation

Knowledge Auto Tagging​

When uploading new Knowledge sources (websites or files), automatic descriptions and appropriate tags are now generated for better knowledge retrieval. This enhances agent context awareness by providing a short summary of what each knowledge source contains. This makes it easier to search, filter, and utilize knowledge in your agent.

---

Enhancements ✨​

Serenity AI Chat​

Serenity AI Chat now delivers more relevant and context-aware conversations by:

• Activating Workbench skills for enhanced capabilities
• Enabling Agentic RAG and Agentic File processing for smarter knowledge retrieval
• Making the chat aware of the current time and user identifier for up-to-date responses

These improvements ensure every user session is more personalized, accurate, and timely.

Email Skill​

The Email Skill now supports AI-generated mode. You can choose between AI-generated or static content. Add custom instructions for how the agent should create the email body and subject, and send agent-customized emails. This allows for more dynamic and personalized communication, tailored to your workflow needs.

Scheduled Triggers​

Scheduled triggers now allow you to target the agent's published version by default. Always run the published version of the agent, avoiding manual configuration whenever changes are made.

---

Bug Fixes & Minor Improvements 🛠️​

• Compatible models now support the "Enable Thinking" parameter configuration.
• Knowledge sources can now be activated or deactivated via API endpoints.
• Chunking strategy has been improved in knowledge processing.
• Knowledge pre-filtering can now be used in all integration modes.
• Improved contrast color ratios and semantic relationships in Chat Widget elements for enhanced accessibility.
• Sensitive data obfuscation has been fixed when the request contained empty text.
• Proxy Agents now support system parameters liquid syntax.
• Email notifications have been added for failed Auto Recharge payments.
• Added tracking for compliance violations with RagMetrics provider in the Compliance Center Dashboard.
• Improve rendering of response output based on model capabilities.
• Fixed execution of Alinia Guardrails when some rules where active.
• Auto recharge is triggered instantly when configuration is set for immediate credit top up.

New Features 🚀​

Agent Clients: Secure authentication​

Agents now support a more secure authentication scheme with Agent Clients. Authenticate using a generated, encrypted JWT token by creating your own client ID, secret, and public key.

This approach enables agents to run more safely, without exposing sensitive API keys in the frontend. For full implementation details refer to Agent Clients documentation

Chat Widget: Enhanced accessibility​

The Chat Widget now offers major accessibility improvements, including a high-contrast mode, improved ARIA labels for screen readers, and design updates that follow UNE-EN 301549 standards.

For full specifications refer to the Chat Widget Accesibilty Documentation

---

Enhancements ✨​

Flexible currency selection​

Billing configuration now allows you to select your preferred currency for displaying and tracking all usage metrics.

Smarter agent import​

When importing agents from a public link, related dependencies such as datasets are now automatically copied for a seamless experience. Features not included in your plan are clearly indicated.

Agent Design Studio API updates​

New API endpoints are available for individual knowledge file management and model parameter configuration. Check out the API Reference for more information:

Improved instance cleanup​

Instance cleanup is now more efficient, reducing storage usage and improving performance in instance grids. You can manually configure log retention in your agent settings if you require longer history.

---

Bug Fixes & Minor Improvements 🛠️​

• Default tenant configurations are now preloaded on first access to the Serenity AI Chat Management screen.
• Fixed an issue that prevented changing the default embedding model for a tenant.
• Resolved a bug where the audit user was not assigned as the last editor when a website was automatically reprocessed.
• Serenity AI Chat agent is no longer available for execution via the Agent Executor Skill.
• Fixed an error when creating a new test plan with a slow connection and rapid form navigation.
• Email notifications, including invoices, are now sent to configured users upon auto-recharge or manual recharge.
• In Agentic RAG and Agentic File Processing integration modes, skill results are now reused for faster responses when inputs are identical.
• General UI adjustments and improvements.
• Fixed an error when exporting a usage chart with duplicate agent names.
• Enabled uploading Volatile Knowledge from base64-encoded files.
• Fixed a multiple tap issue in conversation starters.
• Implemented validation for the maximum number of files in knowledge.
• Fixed a stored XSS vulnerability in the Chat Widget message field.
• Resolved an error in available HTTP methods.
• Improved Chat Widget behavior in full screen mode.
• Minor fixes and typography adjustments.
• Chat History is now copyied to code snippets in Chat Completion Agent
• Allow for Quota alerts to be excluded from tenant admins
• Added copy to clipboard functionality in Skill results in Agent Instance Details

New Features 🚀​

RAG Metrics Guardrails​

Serenity Star now integrates with RAG Metrics as a guardrails provider for agent responses. You can monitor outputs against compliance violations, contextual relevance, and other quality checks before responses are delivered.

This gives teams a clearer way to evaluate response quality and enforce safer, more consistent agent behaviour.

Agentic File Processing for Volatile Knowledge​

Volatile knowledge files now support an agentic file processing mode. Agents can use the current conversation, chat history, and user intent to decide which files are relevant before retrieving content.

By narrowing retrieval to the files that matter most, this improves answer precision while reducing unnecessary token usage.

---

Enhancements ✨​

Refreshed Configuration UX​

Several platform configuration flows now use a clearer card-based experience. Replacing dense dropdowns with guided selection cards makes options easier to understand and quicker to compare.

Improved Error Logs for Skills and Workflows​

Agent Executor Skill and Workflow runs now provide more detailed failure logs. Instance details include clearer error information to help teams diagnose complex execution issues faster.

Workbench Improvements​

Workbench session management has been improved for better performance, file persistence, and usage tracking. A sample prompt with advanced configuration examples is now included when adding a skill, making it easier to get better results from the start.

---

Bug Fixes & Minor Improvements 🛠️​

• Improved validation for file uploads by URL. Invalid URLs now return Bad Request instead of generating a 500 error.
• Fixed a redirect loop when two-factor authentication is enabled.
• Fixed Markdown file uploads in the chat preview in Agent Design Studio.
• Improved code snippets for importing the CSS-customizable Chat Widget.
• Improved several empty states and corrected text across the platform.
• Fixed an issue when duplicating agents that include datasets.
• The HTTP Skill now supports sending a body with DELETE requests.
• Fixed an issue related to token renewal in subscription plans.
• Corrected the lockout icon color in the users table.
• Improved the create-connection side panel so it no longer blocks when creation fails.
• Fixed multiple issues in knowledge loading in the agent creation form.
• Fixed a caching issue when enabling Skills in Serenity AI Chat.
• Fixed an issue where the total token count was not displayed in the chat preview.
• A modal is now shown when users try to access old invoices that are no longer available, prompting them to contact support.
• Added clearer guidance for using CSS in the Chat Widget.
• Added skills to integrate with Fathom.

New Features 🚀​

Agentic RAG​

The new Agentic RAG knowledge integration mode gives agents more control over how they use knowledge sources. Agents can refine search queries, filter documents using metadata tags, and perform multiple search strategies before responding.

This improves context awareness and helps retrieve more relevant sections before the final answer is generated.

Chat Completion Preview​

The new Chat Completion preview in Agent Design Studio makes it easier to test full chat history management. You can edit existing messages, add new ones, and manually shape the conversation history before running the agent.

This gives you a clearer way to understand how chat history affects agent behaviour and how to reproduce that behaviour in your own client implementation.

---

Enhancements ✨​

Agent Subtenant Change​

Agents can now be moved between subtenants more easily. Select the new target subtenant from the General Configuration panel in Agent Design Studio, and the agent will be migrated with the required data.

Connections, API keys, and other configurations tied to the original subtenant are not transferred automatically and must be reconfigured if needed.

Chat Widget​

Specific Agent Version​

You can now specify which version is executed when using the Chat Widget, allowing for better version management and control.

Fully customizable CSS​

We have released a new resource for configuring the Chat Widget with fully customizable CSS classes. This gives you more direct control over the widget's visual style when you want to manage its appearance through your own CSS.

Azure Safety Guardrails Provider​

Azure Safety can now be used as a guardrails provider for your agents, expanding the available options for safety and response control.

---

Bug Fixes & Minor Improvements 🛠️​

• Improved connection creation in Agent Design Studio to prevent the screen from blocking when the process fails.
• Subscription downgrades no longer inactivate AI Chat when it is included in the subscription plan.
• Improved model validation when attempting to use deprecated models.
• Users with the Serenity* AI Chat Management Administrator role scoped to a subtenant can now access audit information.
• Fixed an error in the prompt editor when publishing an older agent version.
• Added the option to show previous conversations in the Chat Widget when conversation starters are configured.
• Improved Chat Widget side panel mode.
• Improved validation in Chat Widget channel configuration.
• Fixed a repeated initial bot message in the Chat Widget.
• Added access to Billing from the Cost Dashboard.
• Improved execution of draft agent versions when changing guardrails configuration.
• Fixed optional postal code handling in billing configuration.
• Embedding generation is now paginated to avoid errors caused by excessive token usage.
• Fixed an issue that prevented custom quota alert emails from being configured.
• Improved billing feedback when adding credit.
• Fixed an error when creating connections for users who have access only to subtenants.
• Improved Microsoft Graph connection refresh behavior.
• Fixed the change-email flow in account settings.
• Added a missing save button in the personal connections form for bearer token authentication.
• Added support for ending WhatsApp conversations with the /EndConversation command.

New Features 🚀​

Serenity* AI Chat​

Serenity* AI Chat has received a major update focused on usability, clarity, and faster access to the tools you use most every day.

Favorite Agents​

Pin your most-used agents and personal assistants for quicker access. Your favorites now appear in the sidebar, making it easier to jump into the right conversation with a single click.

Improved UI​

The chat experience has been refreshed with a cleaner layout, clearer status updates for long-running actions, and better visibility into what the assistant is doing at each step.

Knowledge in Personal Assistants​

You can now enhance personal assistants with your own knowledge base. Add files directly to an assistant and it will use that content for retrieval-augmented generation in its responses.

This helps produce responses that are more relevant to your workflows and content.

Serenity Workbench​

Serenity* AI Chat now integrates with Serenity* Workbench, a dedicated micro-VM environment available directly from your chat conversations. You can run code, edit files, and work in a more capable execution environment without leaving the conversation.

Enable it from Serenity* AI Chat Management.

Connections​

The authentication flow for personal connections in Serenity* AI Chat has been simplified, making it easier to validate accounts, complete authorization, and grant access to the skills your assistants need.

Chat Widget Conversation History​

The Serenity* Chat Widget can now display past conversations, making it easier for users to revisit earlier interactions.

Users can open the widget and review previous messages and conversation history directly from the interface.

---

Enhancements ✨​

Default Subtenants​

You can now configure default subtenants for new users. When this setting is enabled, users who automatically register to your organization through the configured email domain are assigned access to the selected subtenants by default.

Enable default subtenants in the Compliance Center.

Agent Design Studio API​

Agent Design Studio API

The Agent Design Studio API now supports patch methods for granular updates to agent configurations. Update individual agent sections independently:

• Knowledge sources
• Model selection
• Skills (HTTP, MCP, Workflow, Email, and more)
• Behavior and instructions

See the API Reference for details.

---

Bug Fixes & Minor Improvements 🛠️​

• Improved volatile knowledge handling in chat history.
• Allowed chat, copilot, and assistant agents to run in scheduled triggers.
• Improved the UI in the HTTP Skill headers section.
• Improved autosizing behavior across several input areas.
• Fixed an issue where the output JSON format did not update after changing the model.
• Fixed multiple issues in chat widget configuration.
• Updated the Everyone scoped limit so it applies only to logged-in users, not system users.
• Improved the model filter used in the Quality Studio evaluator model selector.
• Improved handling of rate-limit errors in Quality Studio.
• Improved error handling during Quality Studio test-run execution.
• Added support for inactivating datasets in agent knowledge sources.
• Simplified Quality Studio and scheduled trigger creation by inferring the subtenant from the selected agent.
• Tenants now support multiple domains for registration. Contact us for more information.
• Added support for image generation and transcription in Qwen models.
• User administrators assigned to specific subtenants can now grant access only to those subtenants.

New Features 🚀​

Serena Capabilities 🤖​

We have improved Serena's agent creation capabilities. You can now add files directly during the conversation, and Serena will use them as context and knowledge for your agents. Serena can also identify the agent's objective and suggest the necessary skills when needed.

Usage Limits 📊​

Alerts 🔔​

Set custom notifications for your usage thresholds and stay informed before limits are reached. This makes it easier to monitor spending and avoid unexpected interruptions. You can configure your alerts from Usage Limits.

Scoped Limits 🎯​

Scoped limits now provide more granular control over usage across Serenity. You can define individual limits for specific users, groups, user identifiers, or apply them to everyone. You can configure scoped limits from Scoped Limits.

Shared Subtenancy 🤝​

The new shared subtenancy feature makes it easier to create agents that can be shared across your organization without manual subtenant assignment. You can now define whether an agent or entity is restricted to the root level, assigned to a specific subtenant, or shared with everyone.

---

Enhancements ✨​

Groups 👥​

Group details now provide a clearer view of how each group is structured. You can easily see which users belong to a group and whether they come from a subtenant or were added manually.

Knowledge 📚​

Knowledge sources can now be activated and deactivated more easily in Agent Design Studio. This simplifies A/B testing, helps optimize costs, and gives you more control over agent configuration.

Chat Widget 💬​

The chat widget now supports a configurable custom disclaimer. You can enable and manage it directly from the channel configuration.

Billing Information 💳​

The billing section continues to improve with a better interface, clearer messaging, and improved tax configuration.

---

Bug Fixes & Minor Improvements 🛠️​

• Fixed service instance grid reload behavior that caused data to load incorrectly.
• Added markdown component sanitization to prevent access to relative paths.
• Improved API validation errors when invalid routes are accessed without authentication.
• Fixed an issue when selecting an API key on the services screen for users with access to only one subtenant.
• Fixed a display issue that prevented messages from rendering correctly when the transcription service failed.
• Added validation to prevent creating more test cases than allowed.
• Fixed an error when accessing Manage My Account from AI Chat.
• Fixed subscription user counting issues.
• Updated the SDK so chat completion agents use the correct agent version in execute.
• Added validation in Agent Design Studio to prevent selecting the same fallback model as the primary model.
• Fixed scroll issues in several side panels.
• Fixed an error when using shared-type connections in a subtenant scope.
• Fixed an error that occurred when creating an instance if the vendor failed.
• Fixed workflow execution in subtenants.
• Adjusted default embedding model selection logic to use tenant settings or the system default value.
• Improved and expanded empty-state messaging across multiple system screens.
• Applied general styling improvements across several screens.
• Improved performance when loading data on the agent instances screen.
• Increased the allowed field size in the test plan during both import and save.
• Improved feedback when importing a test plan template.
• Adjusted test plan execution with that failed with some evaluator model.s
• Updated the JavaScript SDK to support cancellation tokens and stopping requests during streaming.
• Updated the chat widget to allow canceling message streaming.
• Improved the agent creation API so knowledge configuration can use default values when not specified.
• Updated the agents API to accept either modelId or modelCode plus vendor in model fields.
• Improved vendor validations for AI service execution.
• Optimized instance grid queries.
• Improved feedback and attention notices when accessing agents with deprecated models.
• Increased the character limit for AI Chat grounding.
• Added support for connectors using API key authentication and custom headers.
• Improved knowledge parsing in Hybrid mode.
• Improved the API instances endpoint to return more information.