New Features 🚀​

Microsoft Teams Integration​

Serenity* Star is now available in Microsoft Teams. Install the app from the Microsoft Marketplace, sign in with your Serenity* Star account, and start using your AI agents directly in chats and channels—without leaving Teams.

Key highlights:

• Use AI in Teams – Interact with Serenity* Star agents without leaving Microsoft Teams.
• Multiple conversation contexts – Use agents in personal chats, team channels, and group conversations.
• Trusted knowledge responses – Get answers grounded in your organization's configured knowledge sources.
• Secure access – Sign in with enterprise-ready authentication, including SSO support.

To get started, enable the Teams channel on your agent in the Agent Design Studio, then install the app from the Microsoft Marketplace.

Enhancements ✨​

Security – API Key Management​

We’ve strengthened API key security across the entire platform. API keys are no longer exposed in the DOM, and their rendering, copying, and visualization have been redesigned to minimize the risk of accidental exposure. Clipboard interactions are now handled securely, and sensitive operations are processed through controlled server-side actions. These improvements significantly reduce the surface area for data leaks while maintaining a smooth user experience.

MCP Skills – Tool Discovery Support​

MCP Skills that use authentication schemes now support the Get Tools method. This allows you to retrieve and preview the list of tools available on a remote MCP server before configuring your agent. By making external capabilities more transparent, it becomes easier to understand available actions and fine-tune agent behavior accordingly.

Improved Volatile Knowledge Similarity Search​

Volatile Knowledge within chat history now includes similarity-based retrieval. The system analyzes past messages and dynamically attaches the most relevant fragments. This improves contextual continuity and enhances response accuracy, particularly in longer or more complex conversations.

Agent Design Studio API – Skill Definition Support​

The Agent Design Studio API now allows Skills to be defined programmatically when creating or updating agents. This expands customization possibilities for teams building their own integrations or external clients. Supported API Skills include:

• Agent Executor
• Code Execution
• ImageGenerator
• Prompt
• ChartGeneration
• SpeechGeneration
• WebSearch

and more. Check out the API Reference for full definitions and configuration settings.

Share Directly from Agent Design Studio​

Agents can now be shared directly from the Agent Design Studio view. This simplifies collaboration and reduces friction when distributing agents internally or externally, providing faster access without additional configuration steps.

Billing and Usage Limits Experience​

We’ve refined the billing and usage limits experience to provide clearer visibility into account activity. Usage and rate limits are now presented more transparently across the cost dashboard and billing overview account settings. These changes make it easier to understand active limits and manage your subscription with confidence.

Bug Fixing & Minor Improvements​

• The Agent Execution API now returns detailed cost breakdowns associated with the services used during execution.
• Implemented downgrade logic for the Models & Providers feature, automatically restoring configuration to default values when required by plan changes.
• Improved validation and error handling when creating agents via Serena.
• Added call-to-action buttons redirecting users to the Billing section when a knowledge source cannot be processed due to insufficient balance.
• Improved validation and JSON extraction in Hybrid Knowledge mode, reducing parsing errors and increasing reliability.
• Implemented synchronization and concurrency improvements in knowledge source reprocessing to prevent errors
• Added the ability to search agents by code within the Agents grid.
• Optimized email sending related to deprecated models.
• Added the channel field to embedding instances generated from website sources.
• Improved compatibility of tool usage with Anthropic models that include hyphens in tool names.
• Fixed an issue in the HTTP Request Skill where static body payloads were not executing correctly.
• Resolved a saving error in Models & Providers configuration.
• Fixed an issue in the Workbench Skill where results were reused even when the reuse option was disabled.
• Fixed a scrolling issue in side panels that prevented proper navigation.
• Standardized Identity Server emails and resolved issues in the password recovery flow.
• Corrected various localization inconsistencies and typographical errors across the platform.

New Features 🚀​

AI Services​

Serenity* Star now includes a new suite of integrated AI Services. From image generation to embeddings, you can browse the full catalog of available services and stay tuned for new additions. All AI Services are also available via the API—see the API documentation for details.

Image Generation​

Quickly generate images directly in Serenity* Star by choosing the model that best fits your needs. Describe the image, select a model, and generate results in seconds.

Transcription​

Transcribe audio files directly in Serenity* Star by uploading your audio file and selecting your preferred transcription model. Transcriptions are generated instantly and can be easily copied.

Embedding Generation​

Transform any text into an embedding vector using the embedding generation service to power search, retrieval, and semantic matching use cases.

Improved Cost Dashboard and Instance Tracking​

Access comprehensive audit information, cost and usage dashboard for AI Services within Serenity* Star. The dashboard and instance grid provides clear insights into total usage trends across the entire platform, including costs related to knowledge processing, standalone AI Services, and any internal processes that execute AI Services.

Groups Management​

Create user groups to simplify permission assignment for agents. Groups can be created and updated, then used to grant permissions to agents for all users in the group—improving flexibility, consistency, and ease of administration.

Serenity Workbench Skill​

Serenity* Workbench is a new capability in the ecosystem that allows your agent to access a dedicated workspace for creating, running, and executing code and more. This new skill gives your agent a controlled environment to perform computational and programmatic tasks, complementing LLM behavior with deterministic, code-based workflows.

Enhancements ✨​

Account Settings​

The user account view has been moved to the authentication portal. You can now manage all your account settings directly at https://auth.serenitystar.ai/Account/Manage.

Bug Fixes & Minor Improvements 🛠️​

Agent Design Studio​

• Fixed API errors when including knowledge sources
• Fixed a bug that prevented model parameters from being set when creating an agent with Serena
• Fixed a bug that caused errors when editing agents with Insights after a plan downgrade
• Fixed behavior when adding and completing input parameters
• Improved agent import with datasets to include the default dataset and dataset relationship when within the same tenant/subtenant
• Improved HATEOAS links in API responses
• Fixed a bug that prevented duplicating a skill when importing an agent
• Fixed a bug that broke execution when the Remote MCP skill had no authentication
• Fixed a bug when using Agent Input Parameters within the Prompt Skill
• Skill catalog and documentation now display instructions for connector skills
• Improved authorization access to the Get Models endpoint for all authorized users
• Modified roles for the GetModels endpoint to retrieve all roles via API
• Improved response results in the model endpoint
• Fixed a bug in the preview when attaching a file without sending any text
• Added localization for the timeago chat widget
• Isolated generated HTML scripts for improved security
• Fixed a bug that caused deprecated models to be displayed in the documentation catalog
• Improved dashboard export to exclude grouped information when categorized under "Others"
• Fixed a bug that incorrectly modified the last modification date and last editor when sharing a version
• Improved redirect and confirm links in the authentication site
• Improved loading performance of screens in views with tab navigation

New Features 🚀​

Transcription in Agent Design Studio 🎙️​

The preview section in Agent Design Studio now supports audio recording when an audio model is enabled on the agent.
This allows you to quickly test and iterate on agent behavior using the active audio model configuration.

Embedding Generation Service 🧠​

A new Embedding Generation Service is now available through the API, enabling embedding creation for downstream use cases such as semantic search, similarity matching, and RAG workflows.

Check out the API documentation for more details.

Enhancements ✨​

JavaScript SDK Updates 🧩​

The JavaScript SDK has been updated to improve performance and integration. New capabilities include:

• Transcription support
• Volatile knowledge management

Repository:
https://github.com/serenitystar-ai/serenitystar-js/tree/main/packages/sdk

Chat Widget Improvements 💬​

Transcription 🎧​

The Serenity Chat Widget now fully leverages agent-level transcription capabilities, following the agent’s default configuration to provide a more consistent and reliable audio experience.

File Uploads 📎​

Improved handling of uploaded files in the Chat Widget, with:

• Better alignment to agent configuration
• Improved error handling and validation

User Management 🔐​

User creation, password recovery, and account lockout actions have been updated to use a more secure authentication flow, improving overall account security.

Proxy Agents 🧭​

Proxy agents now support attaching volatile knowledge directly to the system message.
This enables similarity search to be used as part of the system definition, providing a more flexible and effective approach to RAG-style implementations.

{
    "model": "gpt-4o-mini",
    "vendor": 0,
    "messages": [
        {
            "role": "system",
            "content": "You are a helpful assistants",
            "volatileKnowledgeIds": ["<VOLATILE_KNOWLEDGE_ID_1>", "<VOLATILE_KNOWLEDGE_ID_2>"]
        },
        {
            "role": "user",
            "content": "Hi, could you provide more information on how to assemble the new device?"
        }
    ],
    "max_tokens": 4000,
    "stream": false,
}

Better Feedback for End-of-Life Models⏳​

When accessing agents that use models approaching end of life, the deprecation date is now clearly displayed in the overview, along with the configured fallback models, providing better visibility and helping ensure timely model updates.

Bug Fixes & Minor Improvements 🛠️​

• Fixed an issue in the Chat Widget where the page content extraction feature did not update when the page content changed.
• Improved style isolation in the Chat Widget to prevent interference with client page styles.
• Fixed an error when importing agents that include knowledge sources.
• Added a new Agent User role, which can be assigned to operate with permissions within agents.
• Added permissions for users with agent administrator privileges to properly access the instance grid and instance detail views.
• Improved card styles to handle text overflow caused by very long content.
• Automatically select the authentication method in skills connectors when only one option is available.
• Fixed multiple issues related to knowledge deletion (.csv files and other assets) in Agent Designer where the UI was not updating correctly.
• Fixed an issue with dataset uploads in agents when the user only had the agent administrator role or permissions.
• Added summary headers in the instance detail view for chat agents.
• Improved input parameter detection on agent designer
• Added a new endpoint for transcription by passing a file Id instead of a binary file
• Improved security and redaction of sensible data when sharing agents.
• Added resend token functionality when creating password if token is expired.

New Features 🚀​

Volatile Knowledge Strategies​

You can now configure how volatile knowledge files are attached to agents. The following strategies are available:

• Attach Full File: When active, this option will try to attach the full file in the agent context. A validation error will be shown if it exceeds the model context length.

• Similarity Search: When active, the agent will only use the most relevant sections of the attached files.

• Hybrid: When active, the agent will try to attach the full file in the agent context, if it exceeds the context window, a similarity search will be done instead.

An on-demand processing mechanism has also been added. If a file does not have the required processing level for the selected strategy, it will be automatically processed when needed.

Model Discounts​

AI Models can now be used with applied discounts, allowing you to optimize costs and increase usage by selecting models eligible for discounted pricing.

The discounts shown in this image are for presentation purposes only.

Enhancements ✨​

Agent Design Studio API​

New improvements and extensions have been introduced to the Agent Design Studio API, allowing for dataset management and .csv files as knowledge sources.

Learn more in the API Reference

Bugs & Minor Improvements 🐛​

• Fixed styling issues in the password recovery email sent from the backoffice.
• Resolved an issue in the obfuscated data audit within the instance details view that prevented the value dictionary from being displayed.
• Updated the agent executor skill to use version 2 of the API.
• Fixed a console error in the webapp layout related to side panel loading.
• Updated the WhatsApp flow to use the agent’s transcription configuration when available.
• Correctly filtered enabled users in the agent permissions dropdown.
• Fixed a side panel loading issue when clicking View More in the instance details to display token distribution.
• Fixed the deletion of datasets that were left in an error state.
• Fixed an issue where the parameter configuration panel in the preview section was shown even when the agent had no parameters. It is now displayed only when active parameters exist.
• Fixed an issue in Quality Studio executions where changing the agent code caused errors when running a test plan.
• Enabled file uploads via URL through the API.
• Fixed Basic Authentication handling in connectors.
• Increased the timeout in the agent execution skill to allow for longer executions.
• Added a new role for agent usage that can be assigned to agent permissions but has no permissions by default.
• Added the link to volatile knowledge files that use vision as part of the user message for better agent context.
• Fixed an issue with the color picker in the chat widget channel configuration.
• Fixed the import of agents with datasets.
• Fixed an error that occurred when attempting to edit an agent version after the code was changed.
• Improved the interpretation of obfuscated data with special characters.
• Improved validation of .csv files when loading agent knowledge.

New Features 🚀​

Agent Transcription Capabilities​

Agents now include transcription capabilities. Choose which transcription model your agent will use and send audio files as input during regular agent execution. At runtime, the audio file will be transcribed and sent to the agent based on your selected model.

Learn more in the API Reference

CSV Import Wizard as Knowledge Sources​

When uploading CSV files as knowledge sources in your agent, a comprehensive import wizard is presented for a smoother experience. The wizard displays sample rows, table schema, and data types. Get visual and clear confirmation on how the agent will query the data for better responses.

Enhancements ✨​

Serenity* AI Chat​

Serenity* AI Chat has been upgraded to use an industry-standard sign-in and sign-up experience. This improves security and unifies login and registration in the Serenity* Star ecosystem.

Agent Design Studio Through API​

Agent creation through API has been expanded to include the following agent capabilities:

• Knowledge sources for your agent can be managed by specifying files or websites
• Subtenancy support to specify which subtenant the agent will be assigned to

Learn more in the API Reference

Agent Quality Studio​

Test plan results now include a new tab displaying any issues that may have arisen during test plan execution. Whether your balance was insufficient, a missing input parameter was configured, or any other related issue occurred, you can clearly see what the error was and the recommended steps to resolve it.

Volatile Knowledge​

Volatile knowledge has been improved to optimize resources and costs. Chat history and token budget management have been optimized for better agent responses. Clear status and error messages provide better feedback on recommended actions.

Bugs & Minor Improvements 🛠️​

• Enhanced validation error messages in Agent Creation API for clearer feedback
• Fixed file name sanitization in audio endpoint to prevent validation errors caused by unsupported characters from vendors
• Improved liquid parameter validation in skills to prevent duplicate parameter names that would cause skill execution failures
• Added vendor name display on API key cards with new search functionality for easier management
• Expanded AI Chat Management charts to full screen width for better visibility
• Applied visual adjustments across various cards to prevent text overflow
• Added action button tooltips in Test Plans and Quality Studio for improved usability
• Improved subtenancy filter to display the count of applied filters
• Implemented fallback strategy for knowledge sources with failed reprocessing, now using the latest valid version of the site
• Fixed Agent Executor skill bug that occurred when attempting to execute agents with object or array input parameters
• Fixed permission assignment bug that prevented the user dropdown from reloading when assigning permissions to an agent
• Added skill variants and connector skills to catalog
• Fixed bug that prevented changing validation type between blocking and alert modes in guardrails
• Fixed Renew Plan button error when subscription payment fails for a client
• Fixed validation error when deleting datasets that are not assigned to any agent
• Resolved multiple issues related to logout from AI Chat and AI Hub
• Corrected erroneous links in documentation
• Added ability to set a generic quota that automatically applies to all users
• Added conversation ID export functionality in the feedback grid
• Fixed access to workflows feature

New Features 🚀​

Improved Login Flows 🔐​

Serenity* Star now includes a modern, industry-standard sign-in and sign-up experience. This improves security and prepares the platform for expanded authentication capabilities across all Serenity* Star products.

Agent Design through API 🧩​

Agents can now be created and updated directly through the API, enabling smoother integrations with custom clients. This update offers control over both general configuration and behavior, allowing teams to automate agent management and maintain consistency across multiple environments.

Learn more in the API Reference

Transcription Services 🎤​

A new transcription API is available for converting audio files into text using your preferred model. This offers more flexibility in managing audio-based workflows and selecting the balance of accuracy and performance that best fits your use case.

Learn more in the API Reference

Serenity .NET SDKs 🧪​

The Serenity .NET SDKs have been updated for more consistent integration. Developers benefit from clearer abstractions, improved structure, and better alignment with the newest Serenity platform features.

View SDK Documentation | View on GitHub

Enhancements ✨​

Datasets 📊​

Dataset creation now includes an improved wizard for CSV uploads, offering schema and data previews. This gives users better visibility into how their data is interpreted and reduces errors during dataset management.

Connections 🔗​

Manage Your Connections 🗂️​

Connection credentials can now be managed from a single centralized interface. Users can revoke access, set defaults, and maintain full control of integrations with a clearer and more organized experience.

Auth Support 🔑​

Skill connections now offer additional authentication methods. When supported by the selected skill, users can choose between OAuth, bearer token, or username/password authentication. This improves compatibility with external systems and simplifies secure integration.

Improved Knowledge URL Feedback 🌐​

Error handling for URL-based knowledge sources has been strengthened with clearer, more actionable messages. Users now receive specific feedback when a page cannot be accessed or is restricted, making troubleshooting significantly easier.

Bugs & Minor Improvements 🛠️​

• Enhanced input validation in Skills forms
• Improved auto-sizing for initial bot messages in the agent design studio
• Instance IDs are now displayed in grids for feedback, converts, and insights
• Typo corrections and minor UI refinements
• Updated knowledge chunking when token limits are exceeded
• Fixes related to knowledge upload handling in AI Chat
• Return guardrails results in agent execution API
• Fixed issue in minimize chat widget
• Restored editing of model parameters when changing model
• Fixed an error modal that appeared in Agent Design Studio when trying to edit behavior
• Improved embedding generation when a long message was sent, avoiding an error
• Added instance ID to export to Excel in instance grid, feedback insights, and conversation context

🎉 New Features​

📊 Datasets​

Datasets have been improved to make importing, validating, and editing data easier and safer.

Direct Editing: You can now edit datasets directly in the UI to modify records and insert rows without re-importing.

Import Preview & Validation: When adding tables, you get a preview showing row and column counts and the automatically detected schema. Advanced validation highlights type mismatches, missing values, and other common issues before you finalize an import.

Full API Support: Complete CRUDL (Create, Read, Update, Delete, List) API endpoints are available for datasets — see the API reference for endpoint details and usage examples.

🔐 MCP​

MCP integrations now support multiple authentication schemes, including OAuth, API key, Bearer Token, and others.

You can connect to your MCP server by selecting the appropriate authentication type, or reuse an existing integration for faster setup.

📁 Google Skills​

New skills for integrating with Google let agents access files in your Google account with your permission.

Google Drive: Agents can search and reference your documents to access information across your files.

Google Docs & Sheets: Agents can create or edit documents and spreadsheets for report generation, data enrichment, and collaboration.

All file operations (read, search, create, and update) are permission-controlled and require your explicit authorization.

🛠️ Enhancements​

🛡️ Improved Guardrails Message Handling​

Guardrails message handling has been improved to reduce repeated violations from historical messages.

Messages that trigger violations are now replaced with placeholders so past chat history does not re-trigger rules, and violation metadata and context are available in the conversation details pane for faster investigation.

🔁 Replace Knowledge​

You can now replace knowledge sources used by agents without losing associated metadata such as annotations, attribution, tags, and mappings.

Background validation runs during replacement to ensure the new source is compatible before the swap is finalized.

🐞 Bug Fixes & Minor Improvements​

• Improved semaphore handling for message concurrency in conversations.
• Added validation to prevent creating empty datasets.
• Added validation to prevent duplicate dataset identifiers within the same tenant.
• Minor copy and typo fixes across the product.
• Performance improvements for knowledge queries and conversation loading.
• Fixed parameter validation in test execution that prevented creating Quality Studio runs when more than one input parameter was present.
• General text and localization improvements.
• Sidebar improvements for better display on low-resolution screens.
• File type validation on the knowledge upload endpoint.
• Unified styles and validation across login, registration, and password pages.
• General localization fixes and spelling corrections.
• Added a new column in conversations showing the number of messages.
• Added a Models & Providers feature to enable model management (enable/disable models).
• Knowledge side panels now display the file name for context.
• Fixed an initialization bug for boolean-type input parameter fields.
• New version of the Chat Widget with JavaScript style isolation.
• Added validation for tax/fiscal data when configuring auto-recharge.

🎉 Enhancements​

Serenity* AI Chat​

Experience a redesigned Serenity* AI Chat interface built for ease of use. The new UI delivers a cleaner, more intuitive experience:

• Enhanced sidebar navigation — Access all your conversations at a glance directly from the sidebar
• Folder organization — Keep your chats organized by creating custom folders for different projects or topics
• Quick access to company assistants — Instantly find and use pre-configured assistants.
• Flexible model selection — Easily switch between different AI models to match your task requirements

The redesigned interface streamlines your workflow, making it faster to find past conversations, start new chats, and leverage the right AI capabilities for your needs.

API Key Management​

New API Keys can be associated with a virtual user (instead of the key creator), improving auditability and simplyfing access control.

Dynamic HTTP Skill Body​

HTTP Skill now supports dynamic request bodies using context-aware variables. Define a strict template and inject only the fields you need at runtime—useful for parameterized API calls and safer integrations with external services.

Real-time Deobfuscation​

Sensitive fields can be redacted before being sent to the LLM while remaining visible to end users when appropriate. This improves privacy and compliance without degrading user experience.

CSV as Queryable Knowledge​

CSV files in Agent Knowledge sources are treated like queryable datasets. You can filter, search, and run richer queries against CSV-backed knowledge sources to improve retrieval.

Datasets: Replace-and-Append​

New replace-and-append operations make updating agent data safer and simpler — especially for periodic syncs where you want to preserve existing records while adding or replacing rows.

Guardrails & Safety​

Additional guardrail rules have been added to better protect agents from prompt injection, unwanted disclosure of personally identifiable information (PII), and other harmful content categories. These rules improve safety and help enforce organizational policies.

🛠️ UX & Developer Improvements​

Copy-to-Clipboard in Preview Pane​

We added the Copy-to-clipboard for agent responses and user messages in Agent Design Studio's Preview pane.

Message Navigation​

Easily navigate through long messages within a conversation instance using the new message preview.

Agent Name in Navbar​

The agent name is now visible always visible in the Navbar so you can easily identify which agent you are working on.

Input Parameters New Editor​

We completely overhauled the user experience for editing input parameters in the Agent Design Studio. Now, you can easily manage complex input structures with a more intuitive interface, including syntax highlight, auto format and visual feedback for validation errors.

Skill Duplication shortcut​

You can now duplicate an existing skill directly from the skills list.

Bugs and Minor improvements​

• API endpoint to retrieve feedback results.
• Improved the performance of the Compliance Center dashboard.
• In deobfuscated instance details, we now show a dictionary of deobfuscated values and original values so users can map each placeholder to its original value.
• Improved display of deprecation dates when a model is approaching end-of-life.
• Fixed an issue causing manual subscription renewal to fail.
• Fixed an issue in the Quality Studio job that caused test case execution to fail.
• Liquid expressions with assign-type values are no longer mapped as input parameters.
• Improvements to the deletion of expired conversations.
• Restored agent access when the only administrator user was deleted.
• Improved knowledge query performance
• Improved security for sensible connection data
• Changed error messages in Agents Manager and Cost Manager to subtler toast notifications.
• Added validation to prevent creating empty datasets.
• Added validation to prevent creating datasets with duplicate identifiers within the same tenant.
• Added the userIdentifier key to code snippets when this option is enabled.
• Added a quick-access button to instances from the Recent Agents home.
• Fixed translation/language issue for the "view more" text.
• Added an option to expand the user's message in preview.

🎉 New Features​

💬 Feedback Collection​

End users can now provide feedback directly on your Assistant and Copilot agent responses.
Feedback can be submitted through the Chat Widget channel or via API from custom clients.

Gain valuable insights into how users perceive your agents’ performance and identify what aspects they find most helpful.

🔐 Personal Connections in Skills​

Agents can now authenticate to connector skills using Personal Connections.
This enables end users to set up their own credentials and allow the agent to securely interact with their personal accounts — unlocking new types of personalized interactions.

Datasets [Preview]​

Now you can easily upload .csv files to create custom datasets. Link these datasets as Knowledge Sources to your agent, allowing it to reference your data and provide more accurate, data-driven responses.

Want early access?
Be among the first to experience this powerful feature—contact us for early access!

🛠️ Enhancements​

🧩 Context Variables in Skills​

Skills supporting Liquid templating can now use conversation context variables directly. This allows for more dynamic and personalized skill executions based on the ongoing conversation.

💳 Subscription and Invoicing​

⚠️ Downgrade Notifications​

Receive notifications before your subscription plan ends — giving you time to renew or adjust your plan without interruptions.

🧾 Improved Invoicing​

The invoicing system now includes complete fiscal and tax information, along with your billing address.
Additionally, the reverse charge mechanism for corporate customers has been improved for greater accuracy.

Important: For future payments, please ensure that your billing address and tax ID are correctly provided to enable proper invoice generation.

🐞 Bug Fixes & Minor Improvements​

• Conversation variables can now be used in draft versions of agents for easier testing and validation.
• Implemented a retry mechanism in the website reprocessing job.
• Added support for new OpenAI Image v1 models.
• Added a compliance model filter to the AI Chat Management model selection.
• The model end-of-life date is now included in the notification email.
• Adjusted conversation variable display to show all variables instead of only those from the last skill execution.
• Fixed an issue that caused agent duplication when using conversation variables.
• Improved the deprecation date display — when a base model is inactive, its card and deprecation date are no longer shown, and child models can be easily selected.
• Fixed role validation for executing context variable updates.
• Changed the anonymization character used for sensitive data obfuscation.
• Resolved an authentication issue when trying to access the Compliance Center without being logged in.
• Fixed a quota error when upgrading from the Free plan. Now, when switching to any non-free plan, the existing quota is correctly removed.
• Improved performance of the Compliance Center loading process.
• Implemented a session persistence system to prevent users from being logged out after each release.

🎉 New Features​

Conversation Variables​

Introducing Conversation Context and Conversation Variables. Agents can now read, store, and update conversation-level variables during a session. These variables can be injected into prompts, allowing agents to adapt their behavior and outputs as the conversation progresses—enabling more personalized, context-aware interactions.

Define custom variables for the conversation

Access variable data and pass it to the agent in your prompt

Add conditional logic by using the content of the variable

For a practical example of how Conversation Context can be used in a real-world scenario, see our blog post on Identity Verification in Conversational Agents - Context & Flow Control. This article demonstrates implementing an identity verification flow in a customer support agent for online order management, showcasing how conversation variables control multi-step validation and dynamic prompting.

Model End of Life Notifications​

Model Selection​

Easily spot models that are soon to be deprecated directly in the Agent Designer’s model selector, making it simple to plan replacements and avoid disruption.

Email Notifications​

A new email notification service alerts you when an LLM model is approaching end-of-life. Notifications list the affected models, which agents use them, and recommended alternative models, so you can plan migrations and avoid service disruption. Notifications will be sent 30, 7, and 1 day before the End of Life.

🛠️ Enhancements​

Custom Log Retention​

Set how many days advanced instance-level logs are retained per agent in Serenity* Star, giving teams easy control over data-retention policies. Depending on your subscription plan, you can increase the default retention period to better meet your compliance, auditing, or troubleshooting needs, ensuring flexibility and alignment with your organization’s requirements.

Improved API Key Scopes​

Create API keys from Dev Tools and assign explicit scopes to each key. Permissions in the Agent Design Studio are now easier to manage, giving you finer access control for integrations and teams.

Duplicate Workflows​

Duplicate existing workflows and use Serena to quickly modify details. This speeds up iteration, testing, and rollouts for similar workflows.

🐞 Bug Fixes & Minor Improvements​

• Improved loading performance of the billing view.
• Enhanced retry strategies for Knowledge Reprocessing.
• Model selector in agents now shows the correct model types.
• UI improvements, better empty states, and typo fixes.
• Improved usability in the Connection Selector for plugins.
• Fixed data overload in the Duration column in the Agent Instance Grid.
• Improved validation errors in Agent Design Studio for clearer feedback.
• Increased agent size to support more complex agent definitions.
• Improved filtering of allowed models in the model catalog.
• Fixed markdown styles in the Chat Widget.
• Fixed agent duplication bug when changing agent type.
• Improved guardrails by analyzing full chat history in Proxy and Chat agents.