Agentforce is Salesforce’s powerful platform for building and deploying AI agents that help businesses automate tasks, make decisions, and improve customer service. If you are wondering, “What LLM does Agentforce use?” the answer is not just one model; it is a combination of Salesforce’s own custom-built large language models (LLMs) and support for popular third-party options. This setup provides users with flexibility while maintaining security and efficiency. In this guide, we will explain it in simple terms, based on the latest information.<\/p>\n\n\n\n
Before discussing LLMs, let us briefly explain Agentforce. It is part of Salesforce’s Einstein AI suite, launched to enable companies to create “agents,” which are intelligent AI helpers capable of handling complex tasks such as answering questions, processing data, or writing code. These agents use LLMs to understand language, reason, and act. Agentforce stands out because it integrates seamlessly with Salesforce tools like CRM data, making it valuable for sales, service, and development teams.<\/p>\n\n\n\n
Salesforce does not rely solely on external models; they have their own large language models developed by the Salesforce AI Research team. These models are specifically designed for tasks such as coding and business workflows, ensuring high performance and security within the Salesforce ecosystem.<\/p>\n\n\n\n
These homegrown models are the core of Agentforce for Developers (formerly Einstein for Developers), making coding and AI tasks smoother and more reliable.<\/p>\n\n\n\n
Agentforce is not limited to Salesforce’s models. It supports a wide range of managed LLMs from partners such as OpenAI, Anthropic, and Google. These models are ready to use and geo-aware, meaning they comply with local data regulations. Here is a simple table of key options:<\/p>\n\n\n\n Older models like GPT-3.5 Turbo are rerouted to newer versions for better results. This variety lets you choose based on speed, cost, or features. For example, Claude models excel in detailed reasoning, while GPT models are versatile for everyday use.<\/p>\n\n\n\n Want even more control? Agentforce lets you bring your own LLM through integrations with:<\/p>\n\n\n\n This means you can connect custom or preferred models while still using Agentforce’s tools. It is ideal for companies with specific needs, such as enhanced privacy or specialized training. Simply set it up through the Models API, and your agents can use these models for prompts and actions.<\/p>\n\n\n\n Picking an LLM depends on your goals:<\/p>\n\n\n\n Test a few Salesforce doesn’t pick one “best” for everything, so experiment based on your use case. Always check performance metrics like response time and accuracy.<\/p>\n\n\n\n Agentforce does not rely on a single model. The answer to \u201cWhat LLM does Agentforce use?\u201d is a multi-model approach: Salesforce\u2019s proprietary LLMs, such as CodeGen2.5 and xGen-Code, handle coding and business-specific workflows, while third-party options like GPT, Claude, and Gemini support general reasoning, conversations, and creativity. With the Bring Your Own LLM (BYOLLM) feature, businesses can also integrate their own fine-tuned or hosted models. This flexibility ensures that Agentforce can adapt to any workflow while maintaining Salesforce\u2019s high standards for security, trust, and scalability.<\/p>\n\n\n\n No. Agentforce is multi-model. It combines Salesforce\u2019s proprietary LLMs (like CodeGen2.5 and xGen-Code) with third-party models from OpenAI, Anthropic, and Google, plus a Bring Your Own LLM (BYOLLM) option.<\/p>\n\n<\/div>\n<\/div>\n CodeGen2.5<\/strong>: Optimized for coding tasks, fast completions, bug fixes, and Salesforce Apex support. Agentforce supports managed models from leading providers: Yes. With the BYOLLM feature<\/strong>, you can integrate your own or third-party models via Amazon Bedrock, Azure OpenAI, OpenAI, or Google Vertex AI<\/strong>. This is useful if you have specialized or fine-tuned models.<\/p>\n\n<\/div>\n<\/div>\n It depends on the task: Yes. Agentforce lets you configure agents to use different LLMs for different steps. Example: use xGen-Code<\/strong> for generating code, then switch to Claude<\/strong> for reasoning over customer data.<\/p>\n\n<\/div>\n<\/div>\n Yes. Salesforce keeps proprietary and some partner LLMs inside its \u201ctrust boundary\u201d<\/strong>, ensuring enterprise-grade security, compliance, and data privacy.<\/p>\n\n<\/div>\n<\/div>\n Not directly. However, if you host them on supported platforms (like Amazon Bedrock<\/strong> or Google Vertex AI<\/strong>), you can connect them via the BYOLLM integration.<\/p>\n\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n\n\n <\/p>\n","protected":false},"excerpt":{"rendered":" Agentforce is Salesforce’s powerful platform for building and deploying AI agents that help businesses automate tasks, make decisions, and improve customer service. If you are wondering, “What LLM does Agentforce use?” the answer is not just one model; it is a combination of Salesforce’s own custom-built large language models (LLMs) and support for popular third-party […]<\/p>\n","protected":false},"author":26,"featured_media":5869,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-5852","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-salesforce-service"],"rank_math_description":"Wondering what LLM does Agentforce use? Agentforce combines Salesforce\u2019s CodeGen2.5 & xGen-Code with GPT, Claude, Gemini, plus BYOLLM flexibility.","category_names":["Salesforce Service"],"author_name":"Dadich Rami","post_mailing_queue_ids":[],"_links":{"self":[{"href":"https:\/\/differenzforce.com\/blog\/wp-json\/wp\/v2\/posts\/5852","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/differenzforce.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/differenzforce.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/differenzforce.com\/blog\/wp-json\/wp\/v2\/users\/26"}],"replies":[{"embeddable":true,"href":"https:\/\/differenzforce.com\/blog\/wp-json\/wp\/v2\/comments?post=5852"}],"version-history":[{"count":3,"href":"https:\/\/differenzforce.com\/blog\/wp-json\/wp\/v2\/posts\/5852\/revisions"}],"predecessor-version":[{"id":6026,"href":"https:\/\/differenzforce.com\/blog\/wp-json\/wp\/v2\/posts\/5852\/revisions\/6026"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/differenzforce.com\/blog\/wp-json\/wp\/v2\/media\/5869"}],"wp:attachment":[{"href":"https:\/\/differenzforce.com\/blog\/wp-json\/wp\/v2\/media?parent=5852"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/differenzforce.com\/blog\/wp-json\/wp\/v2\/categories?post=5852"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/differenzforce.com\/blog\/wp-json\/wp\/v2\/tags?post=5852"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}Model Provider<\/th> Key Models<\/th> Best For<\/th><\/tr><\/thead> Anthropic (on Amazon Bedrock)<\/td> Claude 3 Haiku, Claude 3.7 Sonnet, Claude Sonnet 4<\/td> Secure, high-trust tasks with Salesforce boundaries<\/td><\/tr> OpenAI \/ Azure OpenAI<\/td> GPT-4o, GPT-4o Mini, GPT-5, GPT-5 Mini<\/td> General chat, reasoning, and fast responses<\/td><\/tr> Google Vertex AI<\/td> Gemini 2.0 Flash, Gemini 2.5 Pro<\/td> Creative tasks and data analysis<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n Bring Your Own LLM (BYOLLM) Feature<\/h2>\n\n\n\n
\n
How to Choose the Right LLM for Agentforce?<\/h2>\n\n\n\n
\n
Conclusion<\/h2>\n\n\n\n
FAQs<\/h2>\n\n\n
Does Agentforce use a single LLM?<\/h3>\n
What are Salesforce\u2019s own LLMs in Agentforce?<\/h3>\n
xGen-Code<\/strong>: Handles both text and code, powering developer assistants and conversational coding help.
These models are built by Salesforce AI Research and tuned for business workflows inside the Salesforce ecosystem.<\/p>\n\n<\/div>\n<\/div>\nWhich third-party LLMs are supported in Agentforce?<\/h3>\n
OpenAI\/Azure OpenAI<\/strong> \u2192 GPT-4o, GPT-4o Mini, GPT-5, GPT-5 Mini
Anthropic (via Amazon Bedrock)<\/strong> \u2192 Claude 3 Haiku, Claude 3.7 Sonnet, Claude Sonnet 4
Google Vertex AI<\/strong> \u2192 Gemini 2.0 Flash, Gemini 2.5 Pro
Older models like GPT-3.5 are automatically rerouted to newer versions for better performance.<\/p>\n\n<\/div>\n<\/div>\nCan I bring my own LLM into Agentforce?<\/h3>\n
How does Agentforce decide which LLM to use?<\/h3>\n
Coding\/dev work<\/strong> \u2192 Defaults to Salesforce proprietary models (CodeGen2.5 or xGen-Code).
General chat\/agents<\/strong> \u2192 Uses GPT or Claude, depending on your configuration.
Custom workflows<\/strong> \u2192 You can prioritize or switch models via Agentforce settings or the Models API.<\/p>\n\n<\/div>\n<\/div>\nCan I switch between LLMs during a workflow?<\/h3>\n
Are Agentforce LLMs secure?<\/h3>\n
Can I use open-source LLMs in Agentforce?<\/h3>\n