Skip to main content

Release 5.0 2025.06.18.1

· 4 min read

✨ New Features​

🤖 Agent Insights​

Agent Insights allow you to automatically analyze past conversations and evaluate specific aspects of agent interactions.

You can define what insights to collect and set the desired coverage. Conversations will then be periodically analyzed to extract the defined data.

alt text alt text

Requires the Agent Insights feature in your subscription plan

Agent Insights analysis incurs a processing cost

🎯 Conversation Objective​

Assistant and Copilot agents now support setting a Conversation Objective, which automatically closes the conversation when the specified goal is met. This helps reduce costs and avoid excessive use of a single conversation.

alt text

🚀 Performance Improvements​

🗄️ Cache Optimization​

We’ve reevaluated and improved cache integration, resulting in smoother operation across all hub layers.

📚 Knowledge​

🧠 Extended Context Knowledge​

The new knowledge integration mode improves how information is organized and retrieved thanks to expanded context windows.

Enable Large Context Knowledge in the advanced settings to access this new behaviour.

Improvements include:

  • Grouped file chunks for better file organization
  • Enhanced metadata for richer context
  • Refined similarity search with extended conversation history analysis for higher knowledge relevance

alt text

🏷️ Custom File Metadata​

You can now define custom metadata for each knowledge file to help the agent assess relevance during conversations.

Files can be filtered using tags by including the knowledgeTags parameter in agent executions:

alt text

curl -X POST \
 -L "https://api.serenitystar.ai/api/v2/agent/<AGENT_CODE>/execute" \
 -H "Content-Type:application/json" \
 -H "X-API-KEY: <YOUR_API_KEY>" \
 -d '[ 
  ...
  {
    "key": "knowledgeTags",
    "value": "Tag1, Tag2, Tag3"
  },
  ...
]'

Knowledge tags work by applying OR logic

🛠️ Enhancements​

🔑 API Standardization of Task Messages​

Server-Sent Event task messages are now standardized using specific keys:

{ LoadingConfiguration, "configuration_load" },
{ ValidatingConfiguration, "configuration_validate" },
{ ValidatingRequest, "request_validate" },
{ AppendingContext, "context_append" },
{ CalculatingTokenBudget, "context_token_budget_calculate" },
{ SearchingRelevantKnowledge, "context_knowledge_search" },
{ SearchingRelevantVolatileKnowledge, "context_volatileknowledge_search" },
{ AskSensibleDataObfuscation, "sensitive_data_system_definition_obfuscate" },
{ InputSensibleDataObfuscation, "sensitive_data_user_input_obfuscate" },
{ CreatingInstance, "agent_instance_create" },
{ UpdatingActivity, "agent_instance_update" },
{ UpdatingChat, "agent_instance_update" },
{ UpdatingProxyRequest, "agent_instance_update" },
{ CalculatingPrices, "agent_instance_fares_create" },
{ ExecutingMetacontrol, "safety_metacontrol_execute" },
{ ExecutingAlinia, "safety_alinia_execute" },
{ InputModeration, "safety_input_moderate" },
{ LoadingPlugins, "skills_load" },
{ ExecutingPlugin, "skills_{0}_execute" },
{ InvokingPluginsWithEmulator, "skills_emulator_execute" },
{ ValidatingOutputJsonFormat, "output_format_validate" },
{ GettingAIResponse, "llm_response" }

🧩 Skills: Active/Inactive Toggle​

Skills can now be toggled as active or inactive to control whether they're included in agent execution. This allows you to experiment without losing skill configuration.

alt text

📝 Markdown Editor in All Agents​

All agent types now support a built-in markdown editor. Markdown rendering performance and the output JSON field display size have been improved.

alt text

🧘 Serenity AI Chat​

Voice interaction​

Use voice commands to send messages to your conversation.

alt text

Automatic Guidelines​

When registering with a corporate email, AI Guidelines are now automatically generated based on your company domain.

🐞 Bugs & Minor Improvements​

  • Exported instance cost added to Quality Studio
  • Improved validation in system definition and behavior
  • Fixed logging error during agent designer validation
  • Excel export in Agent Manager now differentiates input and output tokens
  • Added Forms API endpoint
  • Fixed user usage chart in AI Chat Management
  • Resolved restore issue in chat widget when agent sent audio
  • Added support for application/vnd.api+json header in HTTP request skill
  • Fixed demo page rendering for agents with parameters
  • Fixed skill configuration issues when user had agent permissions but lacked required roles
  • AI Chat Guidelines now generated from email domain on tenant registration
  • Improved isolation of Chat Widget styles
  • Added endpoints to retrieve Forms results and Agent Insight results
  • Fixed executor skill issues when the invoked agent had no input parameters
  • Fixed side panel rendering in agent designer for newly added users with permissions
  • Improved scroll behaviour on Serenity* AI Chat
  • Improved error handling on Serenity* AI Chat
  • Fixed bug in edit message on Serenity* AI Chat