Both are called agents, and they solve different problems. Agentforce is Salesforce's agent layer: conversational, grounded in Salesforce data, configured by an admin, and strongest on customer-facing service and sales work inside the platform. Caddi is back-office automation across systems: you record the task, and it runs as deterministic code over your whole stack, from the shared inbox to the DMS to the billing system.
The cost shapes are different
| Agentforce | Caddi | |
|---|---|---|
| Pricing model | Flex Credits: $500 per 100,000, about $0.10 per action and $0.15 per voice action | Outcome-based, scaling with the work rather than the action count |
| Platform prerequisite | Production grounding generally needs a paid Data Cloud subscription | None. Caddi connects to the systems you already run |
| Who configures it | A Salesforce admin or developer defines the actions and grounding | Whoever does the work today walks Caddi through it |
| Free entry point | Salesforce Foundations includes an allocation of Flex Credits | Pilots start small and scale with usage |
Where each one is stronger
| Agentforce | Caddi | |
|---|---|---|
| Scope of work | Inside Salesforce, grounded in Salesforce data | Across systems: inbox, DMS, CRM, billing, custodians |
| Interaction model | Conversational. The agent chooses actions from the exchange | Unattended. The same steps run on every execution |
| Unstructured input | Strong on text and knowledge grounded in the platform | Built to read varied PDFs and triage shared-inbox email |
| Setup | Admin or developer configures actions, topics and grounding | Record the task once; Caddi writes and maintains the code |
| Best-fit use case | Customer service, sales assistance, in-platform Q&A | Intake, filing, reconciliation, imports, bulk updates |
| Cost driver | Actions consumed, plus the platform underneath | The work completed |
The determinism question
Agentforce is designed to decide: the agent reads the situation and picks which configured action to take. That is exactly right for a conversation, and it is a poor fit for a process where the requirement is that the same thing happens every single time and you can prove it afterwards.
Caddi splits those concerns deliberately. AI is used at setup, to understand the workflow you recorded. Production runs as deterministic code, so the tenth run matches the first, and any model step inside a run is scoped and logged. In a law firm or an RIA, that distinction is usually the whole argument.
Which fits your situation?
Agentforce
Conversational, customer-facing, grounded in Salesforce data. This is Agentforce's home ground.
Related: Agentforce alternatives, before you hire a Salesforce developer, and Salesforce for financial advisors.
Do more with less
See Caddi in action
Tell us where to reach you and the calendar opens right here. In 30 minutes we'll show you how Caddi automates the back-office work that grows with your clients—built, run, and maintained for you.
Frequently asked questions
What is the difference between Caddi and Agentforce?
Agentforce is Salesforce's agent layer for work that happens inside Salesforce, grounded in Salesforce data and configured by a Salesforce admin or developer. Caddi is cross-system automation for the back office: you record a workflow across your DMS, CRM, billing system and inbox, and Caddi runs it as deterministic code. Agentforce makes Salesforce more capable. Caddi runs the work that spans Salesforce and everything else.
How much does Agentforce cost?
Agentforce is priced in Flex Credits: $500 per 100,000 credits, with a standard action consuming 20 credits, so about $0.10 per action and $0.15 for a voice action. Production deployments generally also need a paid Data Cloud subscription for grounding and unified profiles, which is a materially larger line item. Salesforce Foundations includes a free allocation of credits to start.
Do we need Data Cloud for Agentforce?
For limited use, no. For production grounding, unified profiles and retrieval at scale, most buyers end up on a paid Data Cloud subscription. Budget it as part of the platform decision rather than as an add-on discovered later.
Is Agentforce deterministic?
Agentforce actions are configured, but the agent decides which actions to take from the conversation, which is the point of it. Caddi is the other shape: AI is used at setup to understand a recorded workflow, and production runs as deterministic code, the same steps every time, which is what regulated back-office work usually needs.
Should we use Caddi or Agentforce?
Many Salesforce-heavy firms use both. Agentforce fits customer-facing and in-Salesforce work where a conversational agent adds value. Caddi fits the recurring, document-heavy, cross-system work where the requirement is that the same thing happens every time and the run is logged. If your work sits outside Salesforce as much as inside it, start with Caddi.