Skip to main content

Release 5.0 2026.08.25.1

· 4 min read

New Features 🚀​

Long Term Memory [Preview]​

Agents can now retain relevant information across conversations. Enable User or Agent long term memory from the Agent Design Studio: user memory retains context about each individual, while agent memory retains context shared across everyone who interacts with the agent. See Long Term Memory for the available configuration, and My Memories for how each user reviews what an agent stored about them.

Long-term Memory settings with Agent Memory and User Memory enabled, each with its own instructions

This feature is in preview and requires the Long Term Memory feature in your subscription plan.

Skill Approvals​

Sensitive skills can now require explicit approval before they run. When the agent calls a skill marked for approval, execution pauses until the request is approved. See Skill Approvals for the available configuration.

A WebSearch skill call marked Approval required, showing its arguments with Reject and Approve buttons

Connector Policies​

You can now define which connectors are enabled for your company, controlling which external systems your agents and users can reach. Connectors excluded from the policy are unavailable across all agents and the Serenity* AI Chat. You can manage the policy from the Compliance Center.

The Available connectors list in the Compliance Center, with each connector toggled enabled or disabled

Serenity* Apps​

Finance App​

Integrated with Holded, the Finance app provides a dashboard with key business metrics: current balance, expected cash flow to the end of the month, and upcoming expenses and income. An agent analyses the same data and reports the main insights alongside the figures.

The Financial Expert dashboard showing cash position, projected month close, overdue collections and payments, a treasury breakdown and a panel to ask the financial expert

Internal Knowledge Chat App​

The Internal Knowledge app provides a chat interface over your agent, with accompanying links and resources, from internal documents to external references. Answers are grounded only in cited text from your knowledge base, which suits areas such as HR or Quality that answer recurring questions based on written policies.

The Internal Knowledge Assistant, with a searchable list of useful resources beside a question panel offering suggested questions

Media Content Generation App​

Generate content adapted to each platform, including LinkedIn, Twitter and Facebook, from a single brief.

alt text

alt text

alt text

Enhancements ✨​

Bulk Operations on Knowledge​

The Knowledge section now supports bulk operations. Select several sources and reprocess, activate or remove them in a single action.

The Knowledge section with sources selected and the bulk actions Reprocess selected, Activate selected and Remove selected

Users Onboarding​

You can now create users and groups directly from the onboarding flow. Groups are suggested automatically based on your organisation's preferences.

The Add your team onboarding step, with AI-suggested groups above a table for adding users and assigning them to groups

Cancelled Request Detail per Client​

Requests cancelled by the client are now reported in the instance detail, so they can be distinguished from failures when reviewing an execution.

An instance summary reporting that the request was cancelled by the client and the response may be incomplete

Automatic Conversation Names in AI Chat​

Conversations in the Serenity* AI Chat are now named automatically, which makes earlier conversations easier to identify.

Agent Message Feedback via API​

Feedback on agent messages can now be submitted through the API, so quality signals can be collected from your own front end. See Submit feedback for an agent message in the API reference.

Reasoning Configuration in Serenity* AI Chat​

Reasoning detail can now be configured for the Serenity* AI Chat, defining how much of the model's reasoning is exposed to users.

Serenity* AI Chat settings with the Reasoning Detail selector set to Full

Bug Fixes & Minor Improvements 🛠️​

  • Added a search box for the skills already loaded on an agent.
  • The proportional consumption chart tooltips now also show the total.
  • Input parameters created from the system definition are now created as optional.
  • Serena now recommends templates.
  • Fixed an error in the creation of the Trial Page.
  • Improved and isolated the styles of the Pricing component used on the landing page.
  • Fixed an error that prevented certain text documents from being processed (for example, Teams transcriptions downloaded in Word format).
  • Implemented a more robust parser in the data obfuscation module for encoded line breaks.
  • Isolated skills can now be executed by passing an explicit user connection.
  • Chat widget allows for custom status messages to be defined during agent skill execution.
  • Agent Quality Studio now lets you re-run an evaluation or analyse only part of it, so a single change can be verified without repeating a full run.