For more information on these features, click here to download the Technical Release Readiness PDF for September from the Partner Community.

---

Build & Test

Instruction-Only Flex Prompt Templates and Optional Inputs

Create instruction-only prompt templates in Prompt Builder without inputs

Instruction-Only Flex Prompt Templates allow users to create prompt templates in Prompt Builder that do not require inputs. This functionality is particularly useful when context is provided by other resources, such as flows or Apex. Flex templates can accommodate zero to five inputs, offering flexibility in designing prompts tailored to specific business needs.
When configuring inputs, users can determine if each input is mandatory at runtime. The setup includes an option to select or deselect the "Require when template runs" checkbox, enabling templates to execute without certain inputs if desired. This capability enhances the adaptability of prompts, allowing for streamlined operations based on available data.
Flex prompt templates serve as a solution for unique use cases not addressed by existing template types. Users can integrate these templates into various applications, leveraging AI to generate relevant content.
It is advisable for organizations to review their licensing agreements, as some features may require contacting their Salesforce account representative for activation. This ensures that all necessary permissions and configurations are in place for optimal use of the instruction-only templates.

Official Release Note Configuration

Testing Center Onboarding Resources

Access Testing Center Onboarding Resources from the Agentforce Testing Center Setup

This capability is designed to facilitate the onboarding process for new users while providing experienced testers with immediate access to essential reference materials. Users can engage with a guided experience that streamlines the configuration and utilization of the Testing Center’s features, ultimately leading to more efficient testing workflows.
Upon accessing the Testing Center in Setup, users will find an onboarding menu that links to various resources. These include a video demonstration of the no-code workflow, hands-on practice modules in Trailhead, and comprehensive Salesforce Help documentation. This integration ensures that users can quickly familiarize themselves with the Testing Center's functionalities, promoting a smoother initiation into testing processes.
It is important to note that activation of this feature may require the involvement of the Salesforce account representative to address any necessary licensing considerations. This step ensures that users can leverage the full capabilities of the Agentforce Testing Center effectively.

Official Release Note Configuration

---

Setup Your Data

Gain Visibility Into Knowledge Retrieval Quality

Use built-in dashboards and reports to track run-time performance, view trends, and identify areas to improve.

Knowledge/Retrieval Augmented Generation (RAG) Quality Data and Metrics provides a comprehensive framework for monitoring and enhancing the quality of knowledge retrieval in Agentforce solutions. This feature captures quality scores and run-time details related to knowledge retrieval, facilitating the analysis of performance metrics such as Context Precision, Faithfulness, and Answer Relevance. By utilizing built-in dashboards and reports, users can track performance trends and identify areas needing improvement.
The data collected is instrumental for conducting root cause analysis and fine-tuning AI solutions. This feature operates in both production and sandbox environments, allowing for flexible implementation and testing. Users can provision the data model within minutes, with data collection occurring at five-minute intervals and score calculations updated hourly.
To enable this feature, users may need to contact their Salesforce account representative due to potential licensing requirements. Note that turning on data collection increases your org’s credit consumption rate.

Official Release Note Configuration

---

Deploy on Channels

Easily Add Agents to Multiple Channels with Enhanced Connections Setup

The Enhanced Connections Setup in Agentforce Builder allows for streamlined management of an agent’s connected channels. Users can now configure and deploy agents to multiple channels from a centralized Connections panel, which enhances operational efficiency. This feature provides greater control and visibility into each agent's connections, including routing flows, adaptive response formats, and escalation messages.
Connections facilitate the scaling of agent development by enabling the creation of a single agent that can be deployed across various channels. Each connection encompasses components and settings essential for linking an agent to specific channels, such as messaging platforms and email. When users interact with the agent, it utilizes the associated connection to adapt its responses based on the channel's context and the nature of the inquiry.
For organizations looking to implement this feature, it may be necessary for customers to contact their Salesforce account executive to address any licensing requirements associated with the deployment of agents across multiple channels. This ensures that all necessary permissions and configurations are correctly established for optimal functionality.

Official Release Note Configuration

Escalate Complex Employee Conversations to Reps

Escalate complex employee conversations on Experience Cloud to human agents

Agentforce Employee Agent (AEA) has introduced the capability to escalate complex employee conversations on Experience Cloud to human agents. This feature allows agents to transfer sessions to a designated representative or queue when an employee requests assistance or when a conversation is flagged for escalation. The entire conversation history is retained during this transfer, ensuring that context is maintained for the representative taking over.
To utilize this functionality, an escalation topic must be integrated into the employee agent within Agentforce Builder. When connected to an Experience Cloud site using Enhanced Web Chat, the escalation topic can facilitate the transfer based on the defined Omni-Channel flow. It is important to note that while transfers are supported for employee agents on Experience Cloud, agents connected to other platforms such as Lightning Experience or the Salesforce mobile app will not support direct transfers but can be configured to create cases instead.
Organizations may need to contact their Salesforce account representative to enable this feature, as it may require specific licenses based on agent types.
This escalation capability enhances the efficiency of support interactions, allowing for more effective resolution of complex employee inquiries.

Official Release Note Configuration

---

Extend in Ecosystem

Use Amazon Nova Models on the Einstein Platform

Salesforce has integrated Amazon Nova Lite and Amazon Nova Pro as Salesforce-managed models on AWS Bedrock. This feature allows users to utilize these models for specific functionalities, including prompt templates. The integration of these models enhances the capabilities of the Einstein platform, enabling users to leverage advanced generative AI features more effectively.
Customers with appropriate permission sets for Data Cloud and Einstein can access these models through tools such as Prompt Builder, Einstein Studio, and the Models API. This integration supports the development of customized AI solutions tailored to organizational needs.
It is essential for users to be aware that the availability of these models may depend on specific licensing arrangements. Consequently, customers may need to contact their Salesforce account representatives to ensure access to the required add-on licenses.
The addition of Amazon Nova models enriches the ecosystem of large language models available for use within Salesforce applications. Users can benefit from improved performance and flexibility when implementing AI-driven solutions in their workflows.

Official Release Note Configuration

Disable Region Fallback for Azure OpenAI

The capability to disable region fallback for Azure OpenAI requests enhances control over data locality for organizations with stringent compliance requirements. By toggling this setting, administrators can prevent requests that fail on primary model endpoints from automatically falling back to Azure OpenAI services located in the United States. Instead, if a request fails after being retried on all secondary in-region endpoints, an error will be generated, ensuring that all processing remains within the specified regional boundaries.
This feature is particularly relevant for organizations that prioritize data sovereignty and wish to maintain strict adherence to local data regulations. It applies across all generative AI features utilizing Azure OpenAI, including Agentforce, Prompt Builder, and Models API.
While this feature can be enabled directly by administrators in Setup, it is advisable to consult with the Salesforce account executive if any licensing considerations arise. This ensures that the organization fully understands the implications of activating this setting within their operational framework.

Official Release Note Configuration

---

Observe & Analyze

Gain Dynamic Insights with Agentforce Analytics Powered by Tableau Next (Beta)

Agentforce Analytics, now enhanced with Tableau Next, offers a robust framework for evaluating the performance of AI agents. Utilizing the unified Session Tracing Data Model (STDM), this feature meticulously captures and logs every event and interaction within user sessions as distinct events. By clustering and tagging these events, users can efficiently query data through the semantic model, facilitating insights across multiple sessions.
This advanced analytics capability enables organizations to optimize AI agent performance by providing detailed visibility into interactions. Users can analyze agent behavior, identify trends, and assess the efficacy of responses. The integration with Tableau Next enhances the visualization of data, allowing for more intuitive analysis and reporting.
To access this feature, users must ensure they have the necessary configurations and permissions in place.
It may be advisable for customers to contact their Salesforce account executive to confirm any licensing requirements associated with enabling Agentforce Analytics. This feature is positioned as a significant upgrade for existing users, promoting a seamless transition to harness advanced analytics for improved operational efficiency.

Official Release Note Configuration

Explore Insights and Optimize Agent Effectiveness with Agentforce Optimization (Beta)

Agentforce Optimization provides advanced capabilities for exploring and refining AI agents through its integration with Agentforce Session Tracing. This capability records all interactions within user and agent sessions, creating a unified data model that facilitates in-depth analysis.
Users can analyze interactions categorized by shared intents, enhancing their understanding of user engagement and agent performance.
The feature introduces the concept of conversational moments, enabling daily creation and weekly clustering of these moments for granular insights. By leveraging comprehensive session data, users can identify areas for improvement, detect agent hallucinations, and uncover valuable insights that inform optimization strategies. This functionality allows for a more nuanced examination of agent interactions, from initial user requests to final resolutions.
Agentforce Optimization utilizes the Unified Session Tracing Data Model (STDM), which captures every logged event in a session, including individual conversation turns. STDM entities, relevant to analytics and optimization, are provisioned upon STDM enablement in setup. Optimization extends STDM by identifying session moments—sub-sessions representing interactions with a specific user intent. An advanced Large Language Model (LLM) identifies, clusters, and tags the content of these moments.
To utilize Agentforce Optimization, users must have the necessary permissions and access to Data Cloud. It is advisable for customers to consult their Salesforce account executive to ensure they have the required licenses.

Official Release Note Configuration

---

Enterprise Scale, Performance and Trust

Secure Your Agents with Trusted URL Allowlisting

The Trusted URL Allowlisting feature enhances security within the Agentforce platform by enforcing a business's URL allowlist. This capability ensures that agents can only call, generate, or share links that are pre-approved. When an agent attempts to use a URL not included in the trusted list, the link is automatically blocked and replaced with “URL_Redacted.” This mechanism prevents the dissemination of potentially harmful links and helps safeguard sensitive data during agent interactions.
This feature is particularly relevant for workflows that involve external resources, such as agents generating rich content from third-party domains or providing links to external documentation. By requiring URLs to be explicitly allowlisted, organizations can maintain tighter control over the content accessed and shared within Agentforce.
It is essential for administrators to review existing agent workflows for any URLs that may need to be added to the trusted list. Typical Salesforce URLs are allowed by default, but make sure to add any companion org URLs to each org’s trusted URLs. Any URLs that you add to your trusted URLs are allowlisted for use throughout your org, not just in Agentforce. This review process will help ensure compliance with security standards and reduce the risk of unauthorized access to external sites.
Organizations may need to consult their Salesforce account executive for any required add-on licenses to fully utilize this feature.

Documentation Article Configuration

 For more information on these features, click here to download the Technical Release Readiness PDF for August from the Partner Community.

---

Build & Test

Improve Employee Agent Conversation Recommendations

Admin settings to turn welcome and in-conversation suggestions on or off

Customer Need: Agentforce Employee agents now have admin settings to enable or disable welcome and in-conversation suggestions at the agent level, offering more control than the previous always-on activation.

Impact: Agentforce now offers independent control over welcome and in-conversation recommendations, which dynamically provide real-time insights and support during customer interactions. Welcome recommendations appear upon accessing the Agentforce panel from supported record pages, while in-conversation suggestions are presented during active sessions, both aiming to improve resolution times and user experiences. Administrators should be aware that licensing requirements may apply and should consult their Salesforce account representatives.

Official Release Note Configuration

Simplify Access to Employee Agents

Simplify Access to Employee Agents with the New Agent Access Tab

The Agent Access tab in Agentforce Builder enhances the management of Agentforce Employee agent access. Admins can now view, assign, or remove agent access from permission sets and profiles without switching between Setup page. This functionality streamlines the onboarding process and simplifies access control, facilitating faster deployments through a centralized interface.

Official Release Note Configuration

Easily Migrate from Agentforce (Default) to an Employee Agent

New migration flow

Customer Need: Migrate "default" agents to new "Employee" agents.

Impact: The new migration flow efficiently transitions users from Agentforce (Default) to Agentforce Employee agents. This one-click process replicates existing configurations, ensuring continuous agent functionality.

Official Release Note Configuration

Build Employee Agents with the Agent Readiness Checklist

Admin settings to turn welcome and in-conversation suggestions on or off

Customer Need: Organizations need a structured way to ensure complete setup for functional Employee agents.

Impact: The Agent Readiness Checklist minimizes setup inconsistencies and facilitates efficient tracking of progress, ensuring all necessary actions are completed prior to deployment. This leads to the development of fully functional Employee agents and streamlines the deployment process.

Official Release Note Configuration

Generate Test Cases for AI Agents Using Your Unique Data

Generate test case based on the knowledge available to the agent

Customer Need: Organizations need to effectively evaluate AI agent performance, assessing accuracy, quality, and efficiency. This requires tailored, data-driven testing scenarios and scalable solutions for creating multiple test cases.

Solution: This feature allows users to evaluate and enhance AI agent performance by creating data-driven test scenarios. Accessing the Testing Center from Setup or Agentforce Builder, users can define conversation states and context variables. Generating test cases based on knowledge in the Agentforce Data Library allows for leveraging available information and creating multiple test cases simultaneously for extensive testing.

Official Release Note Configuration

More Intuitive Variable Management Experience

Updated Context panel enhances agent variable management

Customer Need: Users require better visibility into where agent variables are used and a streamlined way to manage them for consistent agent behavior and improved security.

Solution: The updated Context panel simplifies tracking variable usage, connects variables to action inputs/outputs during creation, and enhances security and consistency in agent behavior.

Official Release Note Configuration

Verify AI Sources with Citations in Agentforce Testing Center

Quickly verify the sources used by your AI Agent

Customer Need: Users need to verify the accuracy of AI-generated responses and their underlying sources.

Solution: Citations in the Testing Center link directly to source material, improving AI response accuracy and trustworthiness.

Official Release Note Configuration

---

Deploy on Channels

Connect an Employee Agent to Messaging for Web

Customer Need: Organizations need to provide real-time assistance to employees within their portals using AI agents, ensuring appropriate permissions and licenses.

Impact: Integrating Agentforce Employee Agents into Experience Cloud sites enhances employee interaction and service capabilities through MIAW, operating within the logged-in user's context with adherence to the principle of least privilege.

Configuration

Seamlessly Hand Off Conversations to Agentforce Service Agents

Customer Need: Customers need seamless handoffs to Agentforce Service agents without re-entering their questions.

Solution: Agents now retain conversation history, providing a more natural and efficient customer experience and can pick up where the conversation left off.

Official Release Note

---

Extend in Ecosystem

Use Responses From Custom Agent Invocable Actions in Flow Builder

Specify which fields an agent should return as output

Customer Need: Users need to integrate specific, structured outputs from custom agent actions directly into flows.

Solution: This enhances Flow Builder by allowing defined field returns, streamlining workflow automation and data utilization.

Official Release Note Configuration

New Supported LLM Models in Salesforce

Customer Need: UUsers require updated and flexible LLM models for enhanced generative AI capabilities and location-specific adjustments.

Solution: Salesforce upgraded GPT-4o and introduced Gemini 2.5 Flash Lite, GPT 5, and GPT 5 Mini, offering improved performance and customization.

Supported Models

---

Monitor & Analyze

Gain Visibility Into Agent Behavior With Agentforce Session Tracing

Customer Need: Organizations require detailed insights into agent behavior for testing, debugging, and monitoring.

Solution: Agentforce Session Tracing provides comprehensive interaction data, enabling data-driven optimization and improved performance.

Official Release Note Configuration

 For more information on these features, click here to download the Technical Release Readiness PDF for July from the Partner Community.

---

Build & Test

Enhanced Escalation Experience

Customer Need: Previously, if no human agent was available during an escalation, the AI agent would silently exit the conversation, leaving the customer stranded.

Impact: Customers can now continue their conversation with the Agentforce Service agent even if escalation isn’t possible, retaining conversation context. The escalation topic can be customized.

Official Release Note Configuration

Set Customer Expectations About Wait Time

Feature: Allows checking rep availability before escalating a customer conversation using the “Check Rep Availability for Routing” agent action.

Impact: If no reps are available, customers can be provided with an estimated wait time and given the option to join a queue or try again later, improving satisfaction.

Official Release Note Configuration

Enhanced Testing with Conversation History

Feature: Introduces the ability to include past conversation data in test cases within the Testing Center and Testing API.

Benefit: Improves realism and accuracy of testing scenarios by providing a more comprehensive context for evaluating agent performance.

Considerations: Requires Agentforce to be enabled and at least one agent activated; certain licensing may be required.

Official Release Note

Enhanced Agent Response Evaluation

Implements a scoring system (0–5) in the Testing Center to evaluate agent responses for accuracy and quality.

Benefit: Provides a nuanced understanding of agent effectiveness beyond pass/fail, allowing identification of specific failures and informed adjustments.

Considerations: Access may require specific permissions (e.g., Manage AI Agents, Customize Application) and potential add-on licenses.

Official Release Note Configuration

---

Extend in Ecosystem

Use Gemini 2.5 Models

Feature: Adds support for Vertex AI (Google) Gemini 2.5 Pro and Gemini 2.5 Flash as Salesforce-managed models on the Einstein Platform.

Usage: These models can be used for functionality like prompt templates via Prompt Builder, Einstein Studio, and the Models API.

Official Release Note Supported Models

Integration of Anthropic with Agentforce

Feature: Allows selecting Anthropic on Amazon Bedrock as an alternative LLM provider, operating within the Salesforce trust boundary for secure traffic.

Benefit: Optimizes the performance of the Atlas reasoning engine.

Limitations: May have better performance and latency with OpenAI, supports only English, and lacks multimodal features.

Access: Requires contacting a Salesforce account representative for licensing.

Official Release Note Documentation

Geo-Aware Routing for Agentforce & Einstein

Feature: Directs LLM requests to the nearest OpenAI or Azure OpenAI instances based on the Einstein instance’s location.

Benefit: Enhances response times and minimizes latency, especially for international users, by no longer routing all requests through the US.

Details: Uses GPT-4 Omni for reasoning engine calls; US requests go to local OpenAI, other regions to the closest Azure OpenAI.

Access: Should be auto-enabled for most deployments, but some add-on licenses may be required.

Official Release Note Documentation

This round of release updates covers noteable features from April 2025 and May 2025. At the time of this write-up, the release for May was still in progress which means you may not see all of these features in your customer's accounts until early June.

Agentforce Enhancements: Agentforce Employee Agent (AEA) streamlines employee interactions by automating workflows across multiple platforms including Lightning Experience, Slack, and Salesforce Mobile. This is an upgrade to the existing "Default" agent. Other noteable enhancements include new Test Metrics in the Testing API and the ability to connect Agentforce Service Agent to email.

Prompt Builder Enhancements: Prompt Builder now supports multi-modal AI capabilities by allowing users to incorporate files, such as images and PDFs as well as many other new enhancements including new injection and toxicity detection for the Einstein Trust Layer.

Data Cloud Enhancements: Agentforce Data Libraries now support Web Search and also offer real-time progress updates for the build process. Additional enhancements for RAG support have also been released.

MuleSoft Enhancements: MuleSoft MCP Support provides a unified framework for embedding AI capabilities into existing business processes, driving operational efficiency and improved user experiences.

---

Agentforce

Automate Employee Workflows with Agentforce Employee Agent

Agentforce Employee Agent (AEA) streamlines employee interactions by automating workflows across multiple platforms including Lightning Experience, Slack, and Salesforce Mobile. Its natural language interface, coupled with pre-built templates, allows for quick setup and immediate deployment of agents tailored for specific tasks. AEA operates within user context, ensuring responses are role-based and appropriate for the specific user. This capability significantly reduces manual effort when addressing standard topics and actions, thereby accelerating task completion.

The Employee Agent can assist in providing access to company knowledge, performing automated tasks, and streamlining inter-departmental workflows. Its design supports the creation of multiple agents within an organization, overcoming previous limitations of a single agent framework. In addition to enhancing productivity, AEA brings a robust integration with Slack, where agents can be invoked using simple @ mentions. This functionality turns agents into collaborative partners, leveraging conversational context to deliver intelligent responses based on previous dialogues and available data. The framework ensures data security by restricting content access based on user roles and permissions.

Overall, the Agentforce Employee Agent represents a significant enhancement in facilitating employee productivity and efficiency within the Salesforce ecosystem.

Help Link

Improve Agent Quality with New Test Metrics in the Testing API

Enhance your evaluation of agent performance with newly added metrics in the Testing API. This update introduces response-quality metrics such as coherence, completeness, conciseness, and response latency. These metrics provide invaluable insights into agent output, enabling more targeted optimization efforts.

• Coherence evaluates the logical consistency of agent responses.
• Completeness measures whether responses fully address user queries.
• Conciseness assesses the brevity and relevance of responses delivered.
• Response Latency tracks the time taken for an agent to respond, thus impacting user experience.

By leveraging these metrics, teams can identify specific areas for improvement, facilitating a data-driven approach to agent training and performance enhancement. The process for utilizing this feature remains straightforward. Create tests using the Metadata API to configure and deploy across different orgs. Test execution can be performed via the Connect API, allowing for streamlined assessment workflows. This enhancement is designed for optimal use within sandboxes of Lightning Experience, available to organizations with Agentforce enabled and at least one active agent.

Help Link

Connect Agentforce Service Agent to Email

Agentforce Service Agent now supports autonomous email responses, allowing it to directly engage with customer inquiries via email. This feature enables seamless interaction, especially in scenarios like tracking inquiries, where the agent can provide vital information such as estimated delivery dates and tracking numbers directly sourced from the Agentforce Data Libraries or Salesforce Data.

The structure of the email is defined by a customizable template, while the content is driven by topics, actions, and instructions configured for the Agentforce Service Agent. Control over the email communications can be managed via an Omni-Channel flow, ensuring timely responses or utilizing a basic alternative if preferred. Furthermore, activities and metrics related to emails sent by Agentforce can be tracked using a custom report.

These emails will also appear in the case feed, similar to traditional case emails, maintaining the workflow integrity. It is essential to complete preliminary setup tasks for successful activation, including enabling flex credits billing, creating the Agentforce Service Agent, configuring Email-to-Case, and ensuring compliance with messaging standards. When a customer receives an email response from Agentforce, the Sender field displays the designated Email-to-Case routing address. Each email includes a custom legal disclosure indicating that the response was generated using AI, along with an option for customers to escalate their inquiries to a human representative if desired.

This integration not only streamlines communication but enhances operational efficiency in managing customer interactions effectively.

Help Link

---

Prompt Builder

Prompt Builder now supports multi-modal AI with file inputs like images and PDFs, enabling richer outputs for tasks like product descriptions, troubleshooting, and document analysis. The new step-by-step interface simplifies building, testing, and deploying prompts with clear data handling.

Multi-Modal AI with File Inputs in Prompt Builder

Prompt Builder now supports multi-modal AI capabilities by allowing users to incorporate files, such as images and PDFs, directly into their prompt templates. This advancement enables large language models (LLMs) to leverage visual data and document structures alongside traditional text inputs, enhancing the quality and context of the output generated. Key use cases include:

Enriched Product Descriptions: By combining product images with textual metadata, users can automate the generation of compelling product specifications that accurately capture visual elements, improving engagement and search visibility.

Enhanced Troubleshooting: Technical support can utilize visual context from images, such as screenshots, to provide more precise recommendations. For instance, an LLM can interpret error messages in a "Blue Screen of Death" scenario, leading to tailored advice based on the visual input.

PDF Analysis and Comparison: Users can build automation flows that interpret contract PDFs, identifying discrepancies against existing Salesforce records.

File Object and Notes & Attachments: Users can integrate file inputs either through File (ContentDocument) records for technical integrations or via the Notes & Attachments related list for simpler setups (More on how to use it below :point_down: )

This multi-modal capability aims to refine AI responses and enhance workflow efficiency by leveraging rich contextual insights from user inputs, fundamentally transforming how agents and workflows are developed within Salesforce.

Help Link

Step-by-step visuals in Prompt Builder

Salesforce’s new Prompt Builder interface introduces a more intuitive and streamlined experience for creating and managing AI prompts. Designed with a step-by-step visual layout, it guides users through each phase of the prompt-building process, from initial creation to testing and activation. The new interface simplifies the experience, providing clear sections that illustrate how the prompt is constructed, how organizational data is integrated, and what the final response from the large language model (LLM) will look like.

With this update, users can easily ground their prompts in real-time Salesforce data by inserting merge fields using the Insert Resource tool. The Resolved Prompt view allows users to see exactly how test inputs are merged while ensuring that any sensitive organizational data is properly masked before being sent to the LLM. This transparency not only enhances trust but also makes it easier to troubleshoot and refine prompts for better performance.

The updated interface also offers a comprehensive preview of the LLM’s response, helping users evaluate tone, clarity, and accuracy before deploying. Built-in controls like Template & Preview Settings enable users to fine-tune the prompt’s behavior and output. Once satisfied, prompts can be saved and activated for seamless use across Salesforce apps such as Sales Emails and Lightning pages. This new visual and guided approach significantly lowers the barrier to leveraging generative AI, making it accessible even to users with minimal technical background.

Help Link

---

Agentforce and Data Cloud

Agentforce Data Library now supports Web Search, enabling agents to access real-time public information to enhance their responses. Users can easily create a Web Search data library, with progress tracking to know when it’s ready

Agentforce can now use Web Search

With the new capability to integrate web search into Agentforce via the Data Library, agents can now enhance their information retrieval processes. This feature allows agents to perform real-time web searches, enabling them to access a broad spectrum of publicly available data for user inquiries.

To configure a data library for web search, create a new Data Library and select Web Search under Data Type. Once activated, the system automatically generates essential components, such as data streams and a search index, which can be managed within Data Cloud. Monitoring the readiness of the library through the Setup is crucial, as agents can only leverage this functionality once the status updates accordingly.

You can associate this web search data library with existing or new agents. Furthermore, defining a General Web Search topic will enable agents to clarify their capabilities. This topic instructs agents to use only publicly available information and avoid sensitive or personal data retrieval. Important guidelines include asking for clarifications on vague queries, highlighting conflicting information from multiple sources, and ensuring that the information provided is not harmful or biased.

To utilize this feature effectively, remember to remove the existing General FAQ topic, as only one topic can use the Answer Questions with Knowledge action that facilitates this functionality. Once set up, agents are ready to provide enhanced, informed responses powered by real-time web search capabilities.

Help Link

ADL Real-time Updates

Agentforce Data Library now offers real-time progress updates for the build process, giving users clear visibility into when a library is being built and when it’s ready for use.

With clear progress indicators, users are informed when the system is actively constructing the data library. During this time, the assigned agents do not utilize the data from the library in testing or user conversations, ensuring accuracy and preventing reliance on incomplete datasets.

Once the build process is complete, a clear confirmation message informs users that the data library is ready for use. At that point, the library can be tested and deployed with agents in live environments.

This feature helps streamline the setup process and provides assurance that agents are working with fully prepared and accurate data.

Help Link

Use files stored in Salesforce to ground Agentforce (Data Cloud Content Document Ingestion)

Salesforce Data Cloud now supports ingesting and analyzing unstructured data like PDFs and text files, stored as ContentDocument objects in your org. This allows you to transform documents into meaningful insights by indexing them for search, harmonizing them with structured data, and powering AI experiences like generative answers and agent responses.

With this feature, you can go beyond answering questions based solely on knowledge base articles. You can personalize responses for each customer based on documents attached to their account—such as specific contracts, insurance policies, or any other relevant content—making interactions more accurate and context-aware.

Add an Ensemble Retriever to a Prompt Template

Ensemble retrievers allow for sophisticated querying in LLM prompt templates by aggregating data from multiple sources. This feature enhances the grounding of LLM prompts with precise and relevant knowledge, improving the overall response quality.

When utilizing an ensemble retriever, the system runs individual retrievers in parallel, processing each one’s search independently. The results are then combined into a single list which is reranked based on relevance to ensure that only the most pertinent information is returned, optimizing the efficiency of the prompt or agent interaction.

To incorporate an ensemble retriever in Prompt Builder, initiate by creating or editing a prompt template. In the Prompt Template Workspace, insert the retriever by selecting Einstein Search | Ensemble Retriever under the Insert Resources option. This will display the available ensemble retrievers for selection. Customization options enhance the functionality of ensemble retrievers.

Users can define specific search text (agent input) to narrow results, select desired output fields to ground prompts effectively, and set a maximum number of retrieved results.

This capability ensures that prompts leverage the most relevant and contextual insights available, significantly improving the response quality in various applications.

Help Link

Customize Retriever Output with Prefilters

Prefilters in Prompt Builder enhance the specificity and relevance of data returned by the Einstein Search retriever in prompt templates. These prefilters allow technical users to set conditions that constrain the retriever’s search scope, ensuring only pertinent information is fetched.

To configure prefilters, an individual retriever must first be created and activated within Einstein Studio.

By accessing the Template Settings, admins can navigate to the Search Parameters section and specify prefilter criteria. The prefilter input field facilitates the selection of either static values or dynamic merge fields, which will dictate the data returned. Further refinement of the search can occur at the Search Text field, where keywords can be incorporated along with additional merge fields via the resource picker.

In the Output Fields section, users can adjust the fields being returned, tailoring them according to the desired output.

As a final note, changes to retrievers create new versions, and activation is essential for their incorporation into prompt templates.

Help Link

---

Agentforce and Einstein AI + Mulesoft

Mulesoft MCP

MuleSoft now enables the transformation of any application or API into accessible tools and resources for AI agents. This feature leverages the MuleSoft Anypoint Platform to facilitate seamless integration, allowing data from disparate systems to be ingested and utilized by AI agents.

With MCP Support, applications can be registered as MuleSoft connectors, effectively extending the reach of your AI solutions. This allows AI agents to interact with various data sources and triggers, enabling more nuanced and contextual responses during workflows.

By utilizing Visual Builder, developers can create and modify integration flows without extensive coding, thereby accelerating deployment cycles and reducing time-to-value. Hooks into MuleSoft's API-led connectivity approach streamline exposing APIs for AI agents, providing real-time data access and fostering improved decision-making capabilities.

Organizations can utilize AI models and algorithms with real-time transactional data from integrated systems, facilitating advanced analytics and predictive capabilities. Security practices are maintained through MuleSoft’s robust governance features, ensuring that only authorized access is granted to the data being leveraged. This feature supports a broader ecosystem where businesses can enhance customer engagements through intelligent automation and deep understanding of user behavior.

In summary, MuleSoft MCP Support provides a unified framework for embedding AI capabilities into existing business processes, driving operational efficiency and improved user experiences.

Help Link

Agentforce Innovations: New variables, filters, and invocable actions enhance agent control and automation. Faster agent creation with templates and Gen AI, plus OTP verification, Agentforce Service Assistant and AI-powered testing.

Data Cloud Enhancements: Agentforce Data Libraries now index knowledge articles and files using grounding techniques, ensuring responses are accurate, personalized, and directly tied to your organization’s data.

Slack Integration: Agentforce integrates directly into Slack, allowing teams to trigger agents via DMs or mentions. Pre-built Slack Actions handle data retrieval, canvas updates, and workflow management, while Slack Enterprise Search enhances contextual responses.

MuleSoft Enhancements: MuleSoft’s API Catalog now syncs directly into Salesforce, simplifying API management. Topic Center streamlines external system integrations, and the new Agentforce and Einstein AI Connectors support low-code AI agent development and secure, trust-layered interactions.

Expanded Language Support: Agentforce Service Agents now support more languages, enabling localized AI interactions across French, German, Italian, Portuguese, Spanish, and more.

---

Agentforce

Our product team has introduced a set of powerful enhancements to Agentforce, giving developers and admins advanced tools to integrate, automate, test, and manage AI agents across workflows. These updates enable deeper automation within Flows and Apex, robust agent lifecycle management, and faster delivery of reliable, context-aware AI experiences.

Key Enhancements

1. Control Agent Decision-Making with Variables & Control Access to Topics and Actions with Filters: Enhance Salesforce Agentforce capabilities by leveraging context-driven variables for dynamic agent decision-making, and implement conditional filters to secure and optimize topic and action accessibility.
2. Agents in Flow and APEX and Agent API: Integrate Salesforce Agentforce seamlessly into Flows and Apex using custom invocable actions for automated workflows, and leverage the Agent API for flexible, UI-independent generative AI interactions.
3. Customer Verification, Templates, Agent for Setup, Standard Actions and Progress Indicator: Strengthen security with customizable OTP-based customer verification; accelerate agent creation using specialized templates and Gen AI assistance; streamline admin tasks via Agent for Setup; leverage enhanced standard actions; and improve UX with clear progress indicators.
4. Agentforce Testing Center and AI-Generated Test Cases: Accelerate Salesforce agent testing workflows with batch scenario testing via Agentforce Testing Center, and automate comprehensive test case generation using Gen AI for standard and custom objects.

More information below.

---

Enhance Salesforce Agentforce capabilities by leveraging context-driven variables for dynamic agent decision-making, and implement conditional filters to secure and optimize topic and action accessibility.

Control Agent Decision-Making with Variables

Enhance agent autonomy and decision-making by utilizing context, custom conversation, and action output variables within the Salesforce Agent Builder.

• Context Variables: Mapped to object fields, these variables provide read-only access to real-time data during conversations. They're prefixed with $Context and facilitate operations such as maintaining user session context.
• Custom Conversation Variables: Unique to individual agent sessions, these variables store significant data outputs from specific actions, enabling tailored conversational flows. They can be of types: Boolean, Number, or Text.

Practical Usage: Assign a context or custom conversation variable to an action's input for controlled value utilization. Leverage filters to activate topics or actions conditional on variable states. Specify variables in Agent API or topic instructions using formatted references.

To create a custom variable, map an action output to it within the desired topic in the Agent Builder. This capability empowers agents to exhibit refined behavior and adaptability, thereby enhancing the overall customer interaction experience.

Help Link

Control Access to Topics and Actions with Filters

You can now implement filters on topics and actions within your agent's framework to restrict access based on predefined conditions. This feature allows you to ensure that agents interact with certain topics or actions only when specified criteria are met, enhancing security and streamlining user interactions.

With the introduction of filters, agent behavior can now incorporate contextual variables. For instance, you can configure a filter to allow access to a Case Management topic solely when a user's case status is marked as Escalated. Similarly, an agent can facilitate a return request only if the user’s order falls within a specific timeframe, such as the last 30 days.

To set up filters, utilize the Context panel in the Agent Builder to define the required variables. Once created, filters can be selectively applied to specific topics or actions, ensuring that the agent retains strict control over functionality based on user authentication or session variables. This capability not only strengthens your conversational agent's efficacy but also meets compliance and security requirements.

Help Link

---

Integrate Salesforce Agentforce seamlessly into Flows and Apex using custom invocable actions for automated workflows, and leverage the Agent API for flexible, UI-independent generative AI interactions.

Agents in Flows & Apex

Custom agent invocable actions enable seamless integration of Agentforce Service Agents or Default Agents into Salesforce flows and Apex classes, facilitating automation of repetitive tasks. By utilizing these actions, users can invoke agents to execute background operations in response to defined triggers.

When creating a custom invocable action, the admin or developer specifies the task through a sample user message, which the agent analyzes to identify the relevant topics and actions. This setup streamlines the task execution process as the agent operates on the context provided from the flow or Apex class. Additionally, context variables can be defined based on the agent's configurations, allowing for the dynamic passing of essential data.

Multiple invocable actions can be included in a single flow to handle diverse tasks or repeated interactions with the same or different agents. This flexible design enhances workflow efficiency and maintains contextual relevance, ensuring that agents can provide timely and appropriate responses to user requests.

Help Link

Agent API

The Agent API allows developers to harness generative AI by connecting to Agentforce agents programmatically via a REST API. With this capability, users can establish a seamless interaction with AI agents, facilitating tasks across various platforms and workflows without being constrained by UI limitations.

Key Features

• Headless Agent Deployment: Developers can create and deploy agents that function entirely without a user interface, automating backend processes.
• Standardized Interaction: The API defines standardized endpoints, enabling integration with internal systems and third-party applications effortlessly.
• Session Management: Initiate, send messages to, and end sessions with agents programmatically. This includes the necessary API calls to manage session state and interaction.

This feature empowers organizations to build a robust, agent-based ecosystem, where multiple agents can interact in tandem, invoking one another as necessary, thus optimizing the overall workflow.

Help Link

---

Strengthen security with customizable OTP-based customer verification; accelerate agent creation using specialized templates and Gen AI assistance; streamline admin tasks via Agent for Setup; leverage Agentforce Service Assistant; and improve UX with clear progress indicators.

Implement Customer Verification for Enhanced Security

The Customer Verification topic streamlines secure user interaction for Agentforce Service agents by enforcing authentication before sensitive actions. This feature allows developers to specify which topics and actions require user verification, enhancing adherence to security protocols. When triggered, the agent solicits the user's email to send a one-time passcode (OTP) for access control. Upon entering the correct OTP, the agent retrieves the verified user’s context, enabling access to restricted actions like managing reservations or account details. The process is customizable with filters and variable mappings for tailored security measures.

Help Link

Build Specialized Agents with Templates

Agent templates streamline the creation process by providing pre-built configurations tailored to specific business needs, including use-case specific topics, actions, filters, and variables. With this feature, you can create agents such as Agentforce Service Agents or Sales Coaches from existing templates, ensuring consistency and efficiency.

When setting up, users can select an agent template that corresponds to licensed agent types, define the settings, and manage data libraries for context-aware responses. Customization is available post-creation, allowing adjustments in the Agent Builder for optimal functionality.

Help Link

Create and Modify Agents and Topics with Gen AI

Want to get started on your first agent (or add to your collection)? Need a little help? Or are you trying to figure out custom topics? Gen AI is here to assist."

Use Gen AI to rapidly develop custom agents and topics. By providing detailed input, Gen AI generates tailored agents based on your requirements. During agent creation, specify the agent's role and company description. You can also add or modify topics directly in the agent setup. Gen AI streamlines agent creation by offering recommendations for improvements, enhancing the overall capabilities of your agent.

Help Link

Agent for Setup, Your Sidekick for Admin Tasks

Agent for Setup enhances admin efficiency by integrating AI capabilities into everyday tasks. When enabled, it appears in the Agent Builder, allowing customization with topics, actions, and system messages tailored to specific needs.

Key actions include:

• Answer Questions with Salesforce Documentation
• Get and Explain Object Permissions of User
• Identify Field/Object by Name
• Migrate Connected App

Utilize these actions to troubleshoot and enhance the setup process effectively.

Help Link

Agentforce Service Assistant

Agentforce Service Assistant, a Lightning component on the Case record page, enhances case resolution efficiency for service reps. Upon opening a case, it analyzes the feed to generate concise summaries, allowing reps to quickly grasp critical details. With a single click, it provides a tailored, step-by-step service plan based on your organization’s unique data and policies, leveraging AI-grounding techniques. The Assistant utilizes Agentforce topics and instructions to ensure accuracy, improving response times and customer satisfaction without requiring fixed rules or maintenance. Data security is upheld, with no retention of individual case data.

Help Link

Agent Progress Indicator

Agentforce (Default) and Agentforce Service agents now feature distinct progress indicators for each action, improving user trust and transparency. Custom actions can display user-defined loading text, enhancing the interaction experience. Previously, agents utilized a set of generic progress indicators, which lacked specificity.

To implement loading text, access the desired action in Agent Builder and enable the Show loading text for this action setting, inputting a custom string in the Loading Text field. Additional configuration is needed for visibility in Messaging for In-App and Web conversations.

Help Link

---

Accelerate Salesforce agent testing workflows with batch scenario testing via Agentforce Testing Center, and automate comprehensive test case generation using Gen AI for standard and custom objects.

Agentforce Testing Center

Agentforce Testing Center facilitates efficient batch testing of agents in the Agent Builder, allowing for the quick assessment of multiple scenarios simultaneously. By utilizing a CSV upload template, you can prepare and run extensive test cases to evaluate topic and action performance.

Upon execution, the results indicate which test cases passed or failed, enabling rapid identification of issues. Failed cases can be retested directly in Agent Builder, offering detailed logs for further troubleshooting.

This feature supports testing in sandbox environments only, as modifications can impact CRM data. Additionally, users should consider request consumption and limits, including supported languages and geo-aware models based on the Salesforce generative AI platform. Overall, the Testing Center enhances the ability to deliver reliable agents efficiently.

Help Link

Automated Test Case Generation for Salesforce Objects

Gen AI now enables the creation of automated test cases for standard objects (Account, Lead, Opportunity, and Contact) and custom objects, enhancing the development and QA process. Users can provide specific context in the description to tailor the test case generation. For generating test cases related to custom objects, the description should include the custom object name along with relevant sample utterances, formatted as follows: “Generate test cases for [custom object name]. Sample utterances are: [sample utterance], [sample utterance].”

To create test cases for the Answer Questions with Knowledge action, specify the action, citing relevant knowledge articles: “Generate test cases for the Answer Questions with Knowledge action. Use the following knowledge articles for generating tests: [article title], [article title].” This feature streamlines the testing workflow, ensuring comprehensive coverage and effective validation of Salesforce functionalities.

Help Link

---

Agentforce and Data Cloud

Agentforce Data Library Enhancements

Agentforce Data Library enables improved accuracy and personalization in AI responses by indexing knowledge articles and file uploads using grounding techniques. This feature ensures that AI agents provide responses grounded in your organization's specific information, which enhances the reliability of Large Language Model (LLM) outputs.

Data Libraries streamline configurations across Data Cloud and Prompt Builder by automating tasks such as data streaming and search indexing. The segmentation of knowledge sources into chunks optimizes storage and retrieval efficiency, allowing for faster access to relevant information. The indexing process categorizes these chunks, facilitating quicker and more accurate searches by the AI agents.

The retriever component simplifies data extraction from various databases, minimizing manual intervention. Through standard actions such as Answer Questions with Knowledge, agents efficiently utilize indexed data to respond to user inquiries, ultimately improving the overall user experience.

---

Agentforce and Slack

Agentforce now supports direct integration into Slack, facilitating real-time collaboration between employees and digital agents. This integration allows users to engage Agentforce via direct messages or by @ mentioning them in channels, enhancing workflow efficiency without switching platforms. Teams can seamlessly request assistance on tasks such as data retrieval, form completions, and project management directly within their Slack conversations.

The Agentforce Hub serves as a centralized repository, enabling users to search for specific agents based on skill sets, activating relevant agents for a variety of tasks from IT support to marketing strategy.

Additionally, Agentforce now supports pre-built Slack Actions, allowing teams to automate processes such as creating and updating canvases, managing task lists, and constructing workflows. This functionality aims to streamline operations and reduce manual effort in task management. These actions include:

1. Create a Slack Canvas: Facilitates the creation and formatting of a new Slack canvas using provided information.
2. Look Up a Slack User: Enables identification of Slack users by utilizing Salesforce user IDs.
3. Search Slack: Allows for the searching of public Slack messages to retrieve specific information.
4. Send a Slack Direct Message: Enables sending direct messages in Slack, visible in the user-recipient DM.
5. Update a Slack Canvas: Provides the capability to update previously created Slack canvases based on conversation context.

Leveraging Slack Enterprise Search, Agentforce can access and utilize contextual knowledge drawn from conversations and content within Slack. This enables more relevant and precise responses to user queries, tapping into both structured and unstructured data for optimal information delivery. Overall, this integration aims to enhance productivity and facilitate a user-centric approach to digital labor within collaborative environments.

Help Link

---

Agentforce and Einstein AI + Mulesoft

API Catalog

API Catalog provides a unified interface for managing and utilizing APIs within Salesforce, streamlining the automation process. Users can import APIs from the MuleSoft Anypoint Platform by connecting their Salesforce and Anypoint accounts. Once the sync is enabled, API Catalog automatically updates with eligible APIs and their respective versions, allowing users to view detailed specifications and operational statuses.

Users can activate specific API actions for use in Salesforce automations, including flows, Apex code, and agent actions. APIs must conform to specific criteria, such as being RESTful and tagged appropriately in Anypoint Exchange. The catalog also allows for the promotion of API data between Salesforce orgs, ensuring seamless API governance.

Additionally, developers leveraging the Topic Center in Anypoint can enrich their APIs with annotations, enabling better integration with Salesforce agent functionalities. The system supports manual sync restarts and refresh operations to ensure users have the latest data available for automation.

Help Link

Mulesoft Topic Center

Topic Center enhances the integration of external applications within Agentforce by defining 'Topics'—curated sets of actions and guidelines for API interactions. MuleSoft Developers can generate these Topics using the Anypoint Code Builder, allowing them to translate use case requirements from Salesforce Admins into actionable API configurations. This streamlined approach enables Topics to manage actions both within Salesforce and in external systems with controlled behavior.

Once developed, APIs are uploaded to Anypoint Exchange, where governance, security, and access controls are applied. The API Catalog within Salesforce provides a centralized repository for all relevant APIs, facilitating easy access for Salesforce Admins as they configure Agentforce capabilities.

Upcoming enhancements to the Topic Center include testing functionalities during the design phase and a monitoring suite for API asset utilization, set for release in late 2025, further solidifying the synergy between MuleSoft and Salesforce in an era of actionable AI.

Help Link

Agentforce connector

Agentforce Connector enhances MuleSoft's integration capabilities with Salesforce’s Agentforce platform, enabling developers to build and customize AI agents efficiently. This connector supports low-code development environments, allowing for rapid deployment of AI agents without extensive coding.

Key features include authentication with Salesforce orgs and discovery of active agents, providing a streamlined onboarding experience. Developers can initiate sessions with specified AI agent applications, facilitating real-time interaction.

The connector also enables the amplification of Agentforce's functionalities by allowing external application prompts to be integrated into AI agent workflows. This capability fosters innovative orchestration of customer, partner, and employee interactions, enhancing overall engagement and experience.

Help Link

Einstein AI connector

The Einstein AI Connector facilitates integration of Large Language Models (LLMs) with MuleSoft applications through the Salesforce Einstein Trust Layer. With the latest certification by MuleSoft, the open-source connector now follows stringent development and security best practices.

New capabilities include compatibility with Anypoint Code Builder and support for Java 17, enhancing developer experience. The connector can be easily discovered and imported from Anypoint Exchange directly into your Integrated Development Environment (IDE).

Key functionalities of the Einstein AI Connector allow developers to create low-code AI agents and leverage existing investments, such as APIs and templates, for AI agent tooling. All interactions with LLMs are managed securely via the Einstein Trust Layer, while the ability to generate vector embeddings provides essential support for retrieval-augmented generation (RAG) architectures.

Help Link

---

Languages

Agentforce Service Agent in More Languages (Generally Available)

Agentforce Service Agent now supports multiple languages, expanding beyond English and Japanese to include French, German, Italian, Portuguese, and Spanish in certain locales. This is applicable across all enhanced Messaging channels. Language settings can be configured in Agent Builder by specifying a default language and selecting additional options. It is important to note that when operating in a non-English language, system messages will require manual translation. Additionally, the output from some language models may produce variant responses that do not exactly match the designated locale, necessitating careful management of translations and expectations in multilingual interactions.

Help Link

Use More Languages with Prompt Builder

In addition to the languages fully supported by the Einstein Trust Layer, you can use other languages supported by Salesforce. However, the Trust Layer features might not be available for the added languages. To use these languages, turn on Global Language Support from Einstein Setup.

Help Link