Salesforce Models API

 Generations API       - Single-turn interactions

Chat Generations API  - multi-turn interactions

Feedback API          - gathering user feedbacks

Embeddings API        - semantic similarity tasks

Agentforce for Developers

 Salesforce In-House LLMs powering Agentforce for Developers

CodeGen 2.5

xGen-Code

CodeGen 2.5 is a compact model designed primarily for programmer-specific features such as 'code completion' and 'Code Generation'.

xGen-Code, a robust model tailored for Salesforce Developer use cases, excelling in tasks that blend natural language processing (NLP) with programming capabilities.

xGen-Code powers the latest feature, Dev Assistant, and stands out for its ability to handle complex, multi-turn interactions and dynamic chat functionalities, setting a new standard in both open and closed-source model performance.

Chatbots vs Copilots vs Agents

 𝗖𝗵𝗮𝘁𝗯𝗼𝘁 

Focus: conversations and information retrieval. 

Capabilities: Answer FAQs, provide customer support, and collect data. 

Decision-making: Limited and on responses. 

 

𝗖𝗼𝗽𝗶𝗹𝗼𝘁 

Focus: Collaboration and assistance in specific tasks. 

Capabilities: Generate content suggestions, translate languages, answer complex questions, and offer feedback. 

Decision-making: Suggestive, provides options and insights but doesn't make final decisions. 

 

𝗔𝗴𝗲𝗻𝘁𝘀 

Focus: Autonomous learning and action. 

Capabilities: Make decisions, perform tasks, adapt to situations, and interact with the environment. 

Decision-making: Independent, based on learned data and algorithms. 

Public Groups vs Queues

 𝐏𝐮𝐛𝐥𝐢𝐜 𝐆𝐫𝐨𝐮𝐩𝐬: 

->These are collections of users, defined to facilitate data sharing and collaboration. 

->They are ideal for scenarios where a team needs shared visibility and access to records but do not involve assigning ownership of those records. 

->Public groups are manually defined by administrators and can include users, roles, and other groups.

𝐐𝐮𝐞𝐮𝐞𝐬: 

These are collections of records that are not assigned to any specific owner. 

->Queues are used to manage and distribute workloads among team members. 

->Any member of a queue can claim ownership of a record from the queue. 

->Queues are especially useful for managing cases, leads, service contracts, and custom objects.

Winter '25 Release - Admin Highlights

 -> Dynamic Highlight Panel - replace compact layout.Add up to 12 fields,apply filter to fields (also available for mobile)

-> Dynamic Forms available on mobile!

-> Conditional Formatting for fields to show icons and color based on rules you set.

-> Einstein added to Report Formula (boo AI)

-> Custom Report Types (beta) - massively revamped page makes management easier.

-> List views much better performance (moved to lwc instead of aura)

-> Inline Edits to User List View

-> User Access Summary - added how permissions are granted.

-> Object Manager - where permissions are coming from

LWC local dev (Winter 25 Beta)

Instant preview

 -> don't have to deploy changes to org in order to preview your changes.

-> don't have to reload your browser.

Steps required to enable local dev

1.Enable local dev in setup.

2.Install the local Dev plugin in vs code

   sf plugins install @salesforce/plugin-lightning-dev

Einstein Search

 Einstein Search Perform semantic search on structure and unstructured content using Retrieval Augmented Generation.

Semantic search searches for meaning, not keyword.

How to use Data for Semantic search :

-> Select an embedding model

-> prepare the data

-> Convert that data into vectors

-> Store those vectors in a vector database

-> Index everything

-> Create an API for handling the semantic Searches

-> Keep refreshing the vector database as your data changes

Before you can create a retriever, you need to prepare the data.  

Data preparation involves loading, chunking, vectorizing, and storing content in a search-optimized way. 

A search index stores chunked and vectorized data.  When a search index is created in Data Cloud, a default retriever is created automatically.  

Default retrievers can’t be customized.  However, in Einstein Studio within Data Cloud, you can create your own customized retrievers.

Large Action Model (LAM)

 The main difference between a large action model (LAM) and a large language model (LLM) 

is that LAMs can perform actions based on user input, while LLMs can only generate content.

From Large Language Models (LLMs) to the innovative Large action model using xLAM family and xGen-Sales, 

Salesforce is transforming how sales and service teams operate.

Which LAMs Powers Agentforce?

Large Action Model family

➡️ Executes functions

➡️ Triggers real-world actions

➡️ Leaves traditional LLMs in the dust

xGen-Sales and xLAM :

Salesforce AI Research releases xGen-Sales and xLAM AI models to bring humans and autonomous agents together with data and actions.

Agentforce for Developers

 Agentforce for Developers (formerly known as Einstein for Developers) now has new capabilities! 

You can do a multi-turn chat. This means you can ask the assistant to suggest code changes for your requirement, provide more info to steer it in right direction if it does give you satisfactory response.

You can ask it to explain the code. It does a decent job explaining the logic.

You can also ask it to write boiler plate test code and create test cases..

You can ask it to write docs. This is so useful if you need to document your method signatures and parameters and need descriptions if you are building SDKs and libraries.