Software vendor
Salesforce
Salesforce, an NYSE-listed company based in San Francisco, sells Agentforce, its AI agent platform with the Einstein Trust Layer, priced per action or user.
Function
- Ownership
- PublicSourceSalesforce, Inc. Form 10-K for the fiscal year ended January 31, 2026
Salesforce, Inc. common stock (CRM) listed on the New York Stock Exchange, per its Form 10-K for the fiscal year ended January 31, 2026.
- Headquarters
- United StatesSourceSalesforce, Inc. Form 10-K for the fiscal year ended January 31, 2026
- SEC reporting
- YesSourceSalesforce, Inc. Form 10-K for the fiscal year ended January 31, 2026
- Audit firm
- No
- Website
- SourceAgentforce: The AI Agent Platform
- Subprocessor list
- Not stated
- Contract terms
- Not stated
Offering · Platform module
Agentforce
Agentforce is Salesforce's platform for building and running AI agents, which Salesforce describes as autonomous applications that reason through next steps, act within set guardrails and hand complex issues to people. SourceAgentforce: The AI Agent Platform
Salesforce's 10-K says the Agentforce 360 Platform lets companies build workflows, applications and agents with no-code and low-code tools, with its Trust Layer built in. SourceSalesforce, Inc. Form 10-K for the fiscal year ended January 31, 2026
Agents are built in Agent Builder on Salesforce tools such as Flows, MuleSoft connectors and Apex, and come prebuilt for sales, service and HR tasks. SourceAgentforce: The AI Agent Platform
Salesforce's published pricing, October 2026, is by consumption, through Flex Credits per action or per conversation, or per user, under pre-purchase, pre-commit or pay-as-you-go terms. SourceAgentforce pricing
Where it fits
CoE analysis Fits organizations whose customer, service or employee workflows already run on Salesforce data and want agents to act inside those records. SourceAgentforce: The AI Agent Platform
CoE analysis Salesforce's own pages differ on whether prompt masking applies to agents, so masking should be confirmed in the org's Trust Layer settings before agents handle personal data. SourcesTrusted AI: Key PrinciplesMeet the Einstein Trust Layer (Trailhead module unit)
What it does
- Offering type
- Platform moduleSourceSalesforce, Inc. Form 10-K for the fiscal year ended January 31, 2026
- Workflow areas
- SourcesSalesforce, Inc. Form 10-K for the fiscal year ended January 31, 2026Agentforce: The AI Agent Platform
- Agent platform
- Process automation
- Customer size
- Not stated
AI
- Kind of AI
- SourcesAgentforce: The AI Agent PlatformHow Agentforce Works
- Agents and automated actions
- Built into the platform
- Suggests or acts
- ActsSourceAgentforce: The AI Agent Platform
- Availability
- Generally availableSourceAgentforce pricing
Agentforce is sold on a published rate card; no source read uses the words generally available, so ga is an inference from the published pricing. No preview label applies to the platform on the pages read.
- Models named
- Not stated
Control
- Who controls it
- Customer ITSourceHow Agentforce Works
Commercial
- Pricing model
- Usage-basedSourceAgentforce pricing
- Public price
- Not stated
Evidence
- Independent outcomes
- None found in the sources read.
- Sources read
- Vendor-stated 8
- Independent market evidence None read
- Independently verified outcomes None read
- Pricing None read
- Known limitations 1
- Coverage
- Directory record
- Last verified
Security and control
- Single sign-on
- Not stated
Not described on the Agentforce and trust pages read.
- User provisioning
- Not stated
Not described on the pages read.
- Role-based access
- Not stated
Not described for Agentforce on the pages read.
- Administrator controls for AI
- Not stated
Admins and developers set agent instructions, actions and guardrails in Agent Builder. Whether each AI feature can be switched off on its own is not stated.
- Audit logs
- SourcesHow Agentforce WorksMeet the Einstein Trust Layer (Trailhead module unit)
Covers AI actions: Yes
The Audit Trail feature provides data to track AI agent actions and outputs, and Trailhead says it tracks a prompt through each step; storage location, export and retention are not stated on the pages read.
- Certifications
- Not stated
- Data retention
- SourcesTrusted AI: Key PrinciplesHow Agentforce Works
Zero-retention option: Yes
Prompts and outputs: Under zero data retention, prompts and generated responses are never stored by the underlying third-party LLMs. Retention inside Salesforce is not stated on the pages read.
- Data residency
- Not stated
Not stated on the pages read.
Third-party LLM providers are not named.
- Training on customer data
- Not usedSourceTrusted AI: Key Principles
Scope: Third party models
Where stated: Policy
The Trusted AI page, a marketing page, describes zero data retention as a strict policy and says prompts and responses are never used to train the underlying third-party LLMs. Salesforce's own model training is not addressed on the pages read.
AI features
As the vendor or firm describes them. Notes in small print record what the sources leave open.
Agentforce agents and subagents
Autonomous agents for sales, service, HR and other roles with job-specific subagents and prebuilt actions; escalate to human agents when issues exceed their scope.
- Kind
- Agents and automated actions
- Suggests or acts
- Acts
- Who controls it
- Customer IT
- Availability
- Generally available
Control: Configured by Salesforce admins and developers.
SourcesAgentforce: The AI Agent PlatformAgentforce pricingAgent Builder and testing
Admins and developers use natural language to create instructions, actions and guardrails; testing generates test cases and inspects potential responses.
- Kind
- Agents and automated actions
- Suggests or acts
- Not stated
- Who controls it
- Customer IT
- Availability
- Not stated
Control: Salesforce admins and developers.
SourceHow Agentforce WorksEinstein Trust Layer
Dynamic grounding, zero data retention with third-party LLMs, toxicity detection, data masking and an audit trail of agent actions and outputs.
- Kind
- Built into the platform
- Suggests or acts
- Not stated
- Who controls it
- Customer IT
- Availability
- Not stated
Control: Data masking for embedded generative AI features is configured in Einstein Trust Layer setup; masking is described as disabled for agents.
SourcesHow Agentforce WorksTrusted AI: Key PrinciplesMeet the Einstein Trust Layer (Trailhead module unit)
Unclear or conflicting statements
Where the sources are unclear or disagree, both statements are shown with their links.
Whether data masking applies to agents
Statement A
Trusted AI page: data masking replaces sensitive PII or proprietary business data with non-identifiable tokens before the prompt is sent to the LLM.
Statement B
Trailhead unit Meet the Einstein Trust Layer: data masking for LLMs is currently disabled for agents; it is available for embedded generative AI features.
Recorded under D-143; not resolved. The Trusted AI page describes masking in general and does not mention agents; the Trailhead unit excludes agents.
Questions to ask
For this offering
- Is data masking active for our agents today, and if not, what personal data will reach the LLM provider?
- Which LLM providers serve our agents, and where is zero data retention written in our contract?
- How many Flex Credits will a typical agent run consume in our workflows, and what happens at the end of a term with unused credits?
- Which records can an agent create or change without a human step, and where is that limit configured?
- What does the Audit Trail store for each agent action, where is it kept, and for how long?
From the diligence question set
- DQ-06 Does each feature suggest or act? Can it create, post, approve or change records, and where is that limit enforced?
- DQ-07 Who can change each feature's configuration, prompts, thresholds, data connections or model: your company, our administrators or individual users? Can we restrict and log those changes?
- DQ-13 What does the system record for each AI output or action, and can we export and keep those records for as long as we need them?
- DQ-16 Is our data, including inputs, outputs, documents and corrections, used to train or improve any model, yours or a third party's? Is that off by default, and is the commitment in the contract?
- DQ-19 How are the AI features priced: included in the subscription, sold as an add-on or higher tier, or metered by use? What is the billing unit?
- DQ-20 What happens when use exceeds the included amount: overage charges, throttling, or a feature that stops partway through a critical process?