Agentforce Agents
Basic Information
Short description: Agentforce is a service where Salesforce customers can deploy pre-built, customizable AI agents to automate simple business tasks [source]
Intended uses: What does the developer state that the system is intended for?: Applications include shipping management, sales coaching, technical support, appointment management, content creation, and sales development [source]
Date(s) deployed: Made available on October 14, 2024 [source]
Developer
Website: https://www.salesforce.com/uk/?bc=WA
Legal name: Salesforce, Inc [source]
Entity type: Corporation [source]
System Components
Backend model(s): What model(s) are used to power the system?: Agentforce is built on Salesforce's proprietary AI models, including xGen-Sales, xLAM and the Atlas Reasoning Engine [source] [source] [source]
Public model specification: Is there formal documentation on the system’s intend...: Unknown
Description of reasoning, planning, and memory implementation: How does the syst...: Planning and Reasoning: Agentforce uses ReAct prompting, where "the system goes through a loop of reason, act, and observe until a user goal is fulfilled." Memory: The Salesforce data cloud unifies "all customer data and metadata across systems in real time, enabling Agentforce to operate with complete context and precision." [source] [source]
Observation space: What is the system able to observe while 'thinking'?: Agentforce agents operate in the Salesforce ecosystem, inducing Flow, Customer 360, Apex, Datacloud, and MuleSoft (which acts as a "unified solution for integration and APIs") [source] [source] [source]
Action space/tools: What direct actions can the system take?: Agentforce agents take actions in the Salesforce ecosystem, inducing Flow, Customer 360, Apex, Datacloud, and MuleSoft (which acts as a "unified solution for integration and APIs"). [source] [source] [source]
User interface: How do users interact with the system?: "Agent Builder is the low-code builder for Agentforce [...] Setup is simple: Create a job to be done by the Agent by defining topics, giving natural language instructions for that topic, and creating a library of actions for it to choose from. Easily monitor an Agent’s plan of action and test its responses right in Agent Builder." [source]
Development cost and compute: What is known about the development costs?: Unknown
Guardrails & Oversight
Accessibility of components
Weights: Are model parameters available?: While Agentforce is closed source, a 1B quantized version of their backend model is available: "xLAM-1B, specifically, is a non-commercial, open-source model to help advance the science with the research community, while Salesforce uses a much more performant model for Agentforce." [source] [source]
Data: Is data available?: A small subset of the training data is available on a CC-BY-4.0 license [source] [source]
Code: Is code available?: While Agentforce is closed source, a 1B quantized version of their backend model is available: "xLAM-1B, specifically, is a non-commercial, open-source model to help advance the science with the research community, while Salesforce uses a much more performant model for Agentforce." [source] [source]
Documentation: Is documentation available?: Available [source]
Scaffolding: Is system scaffolding available?: Closed source
Controls and guardrails: What notable methods are used to protect against harmfu...: "With Agentforce, teams can use natural language topics and instructions to create guardrails for an agent, including when to escalate or hand off a task to a human. The Einstein Trust Layer enables Agentforce to use any LLM safely by ensuring that no Salesforce data is viewed or retained by 3rd-party model providers." [source]
Monitoring and shutdown procedures: Are there any notable methods or protocols t...: Unknown
Customer and usage restrictions: Are there know-your-customer measures or other ...: None
Evaluation
Notable benchmark evaluations (e.g., on SWE-Bench Verified): Agentforce has not been publicly tested on canonical benchmarks like AgentBench, SWE-Bench, GAIA, and MLE-Bench. However, their backend model temporarily secured first place on the Berkeley Function-Calling Leaderboard [source]
Bespoke testing (e.g., demos): Demo [source]
Safety: Have safety evaluations been conducted by the developers? What were the ...: None
Publicly reported external red-teaming or comparable auditing
Personnel: Who were the red-teamers/auditors?: Internal (Salesforce) and external (unknown).
Scope, scale, access, and methods: What access did red-teamers/auditors have and...: Salesforce subjected their "AI agents to over 8,000 adversarial inputs to pressure-test their boundaries." They also outsourced testing to external vendors, "which simulated attacks on our product using adversarial prompts with the intention of making the product generate biased or toxic outputs." [source] [source]
Findings: What did the red-teamers/auditors conclude?: Limited information about their findings is available at [source].
Ecosystem
Interoperability with other systems: What tools or integrations are available?: Agentforce agents operate in the Salesforce ecosystem, inducing Flow, Customer 360, Apex, Datacloud, and MuleSoft (which acts as a "unified solution for integration and APIs"). Additionally, by default, Agentforce can connect to a variety of messaging platforms like WhatsApp and Messenger [source] [source] [source]
Usage statistics and patterns: Are there any notable observations about usage?: Salesforce reports bringing "5,200 customers live on Agentforce in their sandboxes in the first two days." [source]
Other notes (if any): --