How can we help?

Release Notes I Talkdesk AI Agent Platform
Published • Last Updated

Talkdesk AI Agent Platform™ empowers you to build, manage, and deploy intelligent, collaborative AI Agents that automate tasks, streamline operations, and enhance customer experiences across channels.

 

 

August 3, 2026

Talkdesk AI Agent Platform™ now lets administrators set a maximum number of customer turns per conversation, preventing infinite conversation loops and giving teams more control over session length.

New:

Max Limit Count: Administrators enable Limit Control and configure a Max Limit Count to define the maximum number of customer turns allowed within a session, either as a default across all channels or configured separately for Voice and Digital channels. Only customer messages count toward the limit; AI responses do not. 

Prevents infinite conversation loops and gives administrators direct control over how long an automated conversation can run.

Note: Supported values range from 1 to 50 customer messages. The default value is 20. Disabling the feature removes the limit entirely.

  • Limit Reached Actions: Administrators choose what happens when the configured limit is reached, either ending the conversation automatically or escalating it to a human agent. This allows teams to align the conversation's end behavior with their business requirements and customer experience.

  • Customer Notification Message: Administrators configure the message customers receive when the limit is reached, using the application's default language or adding additional language variations. Ensures customers receive a clear, localized explanation of what happens next.

  • Turn Count System Variable: A new system variable, Turn Count, automatically tracks the number of customer messages throughout the conversation and can be used inside Agentic flows. It enables administrators to build dynamic experiences, such as sending reminders, triggering verification steps, or escalating to a human agent before the maximum limit is reached.

For more information, please read this article: AI Agent Platform: Limit Control - Max Limit Count Configuration.
 

 

[Preview] July 31, 2026 - AI Agent Platform: Contextual Turn Detection

Note: This feature is available in Preview for select customers only.

With the Talkdesk AI Agent Platform™, it is now possible to adjust wait time in real time based on what a contact is saying, instead of applying one fixed pause to every response.

New:

Contextual Turn Detection: AI Agents dynamically adapt their turn-taking behavior based on conversation context, replacing the single global wait-time setting. The agent extends its wait time when it detects the contact needs to recall or read complex information, such as an ID number, an address, a payment card number, or an account number. 

The agent responds quickly to short answers (e.g., "yes/no") and waits longer for structured or multi-part input. This helps to reduce perceived latency on simple responses while preventing interruptions or misheard input when a contact pauses to think or read.

The system extends wait time (0.5s → 3s) when it detects the caller needs to recall or read complex information. Configured at the system prompt level, the LLM classifier triggers an extended timeout for:

  • Personal ID: Full names, dates of birth, driver's license/passport numbers, tax IDs, national IDs.
  • Contact Info: Full mailing addresses, email addresses, phone numbers.
  • Financial Info: Card numbers, bank account/routing numbers, PINs/verification codes
  • Account Details: Customer IDs, order numbers, confirmation codes, policy numbers, membership/loyalty numbers.
  • Complex/Sequential Info: Detailed problem descriptions, multi-step data, anything requiring the user to check physical documents.

 

 

July 30, 2026 - AI Agent Platform: LLM-Driven Simulator for AI Agent Testing

Talkdesk AI Agent Platform™ now features an enhancement to its Simulator.

Enhancement:

  • LLM-Driven Simulator for AI Agent Testing: Builders now define prompts to drive scenario generation within the AI Agent Simulator, configurable for diverse personas, languages, and intents. During implementation, admins and builders configure prompts and run simulations to validate AI Agent behavior across edge cases before publishing. This speeds up testing and troubleshooting by replacing manually defined test scenarios with realistic, prompt-driven inputs.

Note: This enhancement is only supported for chat simulations and requires the AI Agent Platform. No separate activation is required.

For more information on this feature, please read our article on AI Agent Platform's built-in Simulator.


 

July 24, 2026 - AI Agent Platform: Multi-Language Support

The Talkdesk AI Agent Platform™ admin interface now supports multiple languages, bringing it to full parity with the rest of the Talkdesk product suite. 

All configuration screens, labels, tooltips, and validation messages now respect the language preference set in the user's Talkdesk account.

What's New

The AI Agent Platform has adopted Talkdesk's Polyglot localization framework. Previously, the entire admin UI was displayed in English only, regardless of the language configured in a user's account. 

Now, Admins in non-English-speaking markets can configure and manage AI agents in their selected account language, vastly improving the experience.

The following elements are now fully localized:

  • Navigation labels and menu items
  • Field names and configuration screen text
  • Tooltips and inline help text
  • Validation and error messages
  • AI Agent Platform Simulator

This is available for all Talkdesk global customers using AI Agent Platform.

How It Works

No action is required. The AI Agent Platform admin UI automatically displays in the language already configured in the user's Talkdesk account. To update your language preference, go to Account Settings > Language.

 

 

July 2, 2026 - AI Agent Platform: Simulator 

Talkdesk AI Agent Platform™ now includes a built-in Simulator, letting teams test orchestration behavior, routing logic, and responses directly from the canvas before publishing a new version.

New:

Simulator: Users open an orchestration in AI Agent Orchestration and click Test to launch the Simulator, which replicates real conversation flows through both chat and voice channels. Value: Teams validate agent behavior and routing logic before changes go live, reducing the risk of unexpected behavior in production.

  • Chat and Voice Testing Channels: Choose between a Chat simulation, which opens a live conversation window immediately, or a Voice simulation, which offers configurable settings such as call audio quality, language, voice persona, End of Speech Sensitivity, speaking speed, and a phrase list.

  • Simulation Variables: Define Application Input variables (context data normally injected by an external application) before starting a simulation. 

  • Insights and Thought Process: Each agent response includes an Insight indicator, showing the tool calls made during the response and, through View Thoughts, the internal reasoning behind routing decisions.

Note: This feature is only supported for chat simulations and requires the AI Agent Platform. No separate activation is required.

For more information on this feature, please read our article on AI Agent Platform's built-in Simulator.

 

June 11, 2026 - AI Markers for AI Agents

Starting on Thursday, June 11th, 2026, AI Markers will be available for AI Agents experience, extending conversation marker observability beyond Autopilot intent-based experiences. The following capabilities are now available. 

New:

  • AI Markers for AI Agents: AI Markers allow you to define specific points within automations where data can be extracted and monitored. Previously available only for Autopilot intent-based experiences, AI Markers can now also be configured within AI Agent workflows.
  • Supported Workflow Components: AI Markers are available in the following AI Agent workflow components:
    • Execute Action
    • Function
    • End Automation
  • Conversation Observability: Each time one of these components is triggered, the corresponding marker is activated and recorded. This enables customers to track key workflow execution points and gain greater visibility into how AI Agent conversations progress.
  • Reporting in Talkdesk Explore: AI Marker activity can be monitored through the default AI Markers report in Talkdesk Explore, providing visibility into marker activations across Autopilot experiences.

These updates provide a consistent observability experience across Talkdesk AI products, enabling customers to monitor and analyze conversational automation activity from a single reporting framework.

Important Information:

  • To support AI Agent experiences, the existing AI Conversation Markers report has been updated and renamed to AI Markers. The report also includes renamed columns and additional fields.
  • Customers using integrations, automations, or external systems that reference the report name, column names, or column order should review their implementations and make any necessary updates.
  • For more information about the reporting changes, please refer to this documentation: Release Notes | Talkdesk Explore – Knowledge Base

 

[Preview] May 12, 2026 - AI Agent Observability

Important Note: This feature is available in Preview for select customers. During the Preview phase, AI Agent Evaluation is accessed inside the AI Agent Platform. At General Availability (GA), the experience will move to CXA Operations Center (formerly AI Trainer).

On Tuesday, May 12, 2026, Talkdesk will release the following preview feature for AI Agent Platform (at General Availability stage, the experience will move to Talkdesk CXA Operations Center:

New:

AI Agent Observability: Admins and Supervisors can now review previous AI Agent interactions through the new Session History view, gaining end-to-end visibility into how AI agents handle customer conversations. This capability helps teams validate agent behavior, audit performance, and make informed training decisions based on real interactions. AI Agent Observability is accessible from AI Trainer and from the CXA Operations Center. 

What's included:

  • Session History view: Browse a complete log of AI Agent interactions with key details at a glance — contact and channel, Agent Orchestrator, date/time and duration, end-of-automation outcome, and error count. Click Details to drill into any session.
  • Advanced filtering: Quickly narrow results by date range, channel (e.g., Voice, Webchat, WhatsApp, Digital Connect), Agent Orchestrator, end-of-automation status (Completed, Escalated, Timeout, Error, Abandoned), and Interaction ID.
  • Session details with Insights and Errors: Inspect the full conversation turn-by-turn, including workflow used, inputs and outputs, and a dedicated Errors view to support troubleshooting.
  • Full Chain of Thought: Open the Thoughts panel to see the AI Agent's step-by-step reasoning — from user message to agent decisions, skill calls, inputs, and outputs — to better understand how outcomes are produced.

Note: Following the Preview release, more capabilities will be progressively rolled out to General Availability.

For more information, please read our documentation for AI Agent Observability.

 

[Preview] May 8, 2026 – AI Agent Evaluation

Important Note: This feature is available in Preview for select customers. During the Preview phase, AI Agent Evaluation is accessed inside the AI Agent Platform. At General Availability (GA), the experience will move to CXA Operations Center (formerly AI Trainer).

On Friday, May 8th, 2026, we will begin the progressive rollout of AI Agent Evaluation for Talkdesk AI Agent Platform™.

New:

AI Agent Evaluation introduces a batch-testing and regression safety net for AI Agents, enabling AI Agent Admins and technical users to validate agent behavior against curated datasets before promoting changes to production. The following new capabilities are now available:

  • Evaluation Wizard: A guided, multi-step flow to configure and launch an evaluation run. 
  • Manage Datasets: A centralized library to manage evaluation datasets. Two dataset types are supported: Scripted Datasets and Dynamic Datasets.
  • CSV Upload & Download: Create datasets via the UI form, or bulk-import them via CSV. Existing datasets can be downloaded as CSV.
  • Metrics Library: Configure the metrics that matter to your use case, including Answer Accuracy, Tool Correctness, Goal Accuracy, Instruction Adherence, and Guardrails.
  • Evaluation List & Results: Review all past evaluation runs in a single list view, with the dataset used, the agent version evaluated, and run status at a glance.
  • Detailed Scenarios: Per-scenario Pass/Fail results against each configured metric, with red/green indicators so regressions stand out immediately.
  • Session Drill-Down: Inspect any scenario turn by turn, full chain of thought, tool calls, metric scores, and the reason for failure for each metric.
  • Re-run Evaluations: Re-execute a completed or failed run against a different agent version or dataset, making it easy to validate results across the non-deterministic nature of LLM-based evaluation.

Resources

For more information, please check our documentation for AI Agent Evaluation. Also, you can see the demo here.

 

April 28, 2026

On Tuesday, April 28th, 2026, we are releasing new capabilities for the Orchestrator Configuration panel within the Talkdesk AI Agents Platform™. All features described are available in all accounts, but must be manually enabled and configured for each AI Agent individually.

What’s New

  • Custom Welcome Message (Multi-language and multi-channel support): The Custom Welcome Message now supports per-channel and per-language configurations. Admins can define distinct greeting text for all channels, including Voice, Chat, SMS, and WhatsApp, ensuring a localized and channel-specific experience for every contact.
     
  • Inactivity Message: When a contact becomes silent during a voice call, the Agent can now automatically send a follow-up message. This re-engages customers automatically by allowing Admins to configure silence thresholds and retry attempts. Note: This feature is available for the Voice channel only. 
     
  • Dynamic Contextual Fillers: AI Agents can now automatically generate short filler messages during voice calls when a response is taking longer than expected. Messages are AI generated in the contact's language and tailored to the active Skill and conversation context. Note: This feature is available for the Voice channel only. 
     
  • Guardrail Customization: Admins can now define the exact message sent to contacts when a guardrail is triggered. This replaces platform defaults with brand-aligned copy across all channels and languages.

Resources: 

For more information, see:  

  • AI Agent Platform: Settings - Custom Welcome Message in this article.
  • AI Agent Platform: Inactivity Message Configuration in this article. 
  • AI Agent Platform: Settings - Conversation Fillers in this article. 
  • AI Agent Platform: Guardrail Customization in this article. 
     

 

April 13, 2026

On April 13th, 2026, the following new feature will be available for all customers.

New:

  • Model Context Protocol (MCP) for AI Agents is now Generally Available: Talkdesk now supports MCP integration, enabling customers to extend AI Agents with externally hosted tools. 
    • With MCP skills, AI Agents can securely invoke remote tools exposed by customer-hosted MCP Servers.

For more information, please check our MCP documentation.

 

[Preview] February 2, 2026

Important Note: This feature is available in Preview for select customers.

On February 2nd, 2026, the following new feature will be available in Preview. 

New:

  • Model Context Protocol (MCP) for AI Agents: Talkdesk now supports MCP integration, enabling our customers to extend AI Agents with externally hosted tools. With MCP, AI Agents can securely invoke remote tools exposed by customer-hosted MCP Servers.

For more information, please check our MCP documentation.

 

 

December 18, 2025

On Tuesday, December 16th, 2025, Orchestrator configurability will be available in AI Agent Platform™, with the following new capabilities:

New:

  • Orchestrator Configuration Interface: A dedicated configuration panel that allows teams to control how the AI Agent greets contacts and handles pauses or longer processing times across supported channels.
    • Custom Welcome Message: Define a greeting that is sent at the start of every conversation. This ensures a consistent and fast welcome that does not rely on LLM generation.
    • Conversation Fillers: Configure predefined short messages that keep the interaction active when the system needs more time to respond. You can set timing, retry limits, and provide separate initial and retry messages. This feature is currently supported in the voice channel.

For more information, please check our documentation.

 

July 31, 2025

On Thursday, July 31st, 2025, Talkdesk AI Agent Platform™ will become Generally Available (GA) to Talkdesk customers who wish to purchase access to it.

Talkdesk enables organizations to create and orchestrate intelligent AI Agents that work collaboratively to handle tasks, optimize operations, and deliver seamless customer experiences across channels.

With this full release, businesses can configure multi-agent workflows, where AI Agents coordinate, share information, and perform specialized actions as part of a broader automation strategy. AI Agent Platform’s modular design enables more robust and flexible automation across your business.

What’s included

This release includes the following main features:

  • Multi-Agent Orchestration.
  • Knowledge Management Integration.
  • Custom Guardrails.
  • Advanced Workflow Automation.
  • Visual No-Code Builder.
  • Native Talkdesk Integrations.
  • Scalable Architecture.

For more details on each of these features, please refer to this article section of our Knowledge Base. 

How to Access / Install

If you wish to start using the AI Agent Platform, please reach out to your Talkdesk representative.  

For more information, please check our AI Agent Platform documentation.

 

 

[Preview] June 30, 2025

Important Note: This feature is available in Preview for select customers.

On Monday, June 30th, 2025, we will release Talkdesk AI Agent Platform™.

AI Agent Platform empowers you to build, manage, and deploy intelligent, collaborative AI Agents that automate tasks, streamline operations, and enhance customer experiences across channels.

Now, you can create a single AI Agent or a multi-agent orchestration, where multiple AI Agents work together, communicating and delegating responsibilities to one another, with each contributing their specialized skills to complete parts of a larger objective. This modular design enables more robust and flexible automation across your business.

 

What's Included

  • Multi-Agent Design: Build individual AI Agents with distinct responsibilities and define how they interact.
  • Knowledge Management integration: AI Agents can connect to Talkdesk Knowledge Management™ (KM) to search for relevant information and improve replies to customers.
  • Custom Guardrails: Define guardrails to control and refine the AI Agent’s responses before they are delivered to the customer, ensuring tone, accuracy, and appropriateness.
  • Advanced Automation - Workflows: Use workflow logic to handle branching paths, conditional logic, and dynamic decisions, all without human intervention.
  • No-Code Interface: Design AI Agent behaviors, flows, and skills visually through a node interface. This allows both technical and non-technical users to build and maintain AI Agents with ease.
  • Seamless Integration with Talkdesk: Integrate AI Agents effortlessly with existing Talkdesk products to extend automation and intelligence across your workflows.
  • Easy Scalability: Expand your AI Agent system by adding new skills to handle a wider range of tasks and automations.

How to Access

For more information, please reach out to your Talkdesk representative or Customer Success Manager.  

All Articles ""
Please sign in to submit a request.

Related to