Agentforce Pricing Explained: Flex Credits, Conversations, Users, and Hidden Costs

Salesforce publishes a tidy headline number for Agentforce, and it is easy to quote. The reality is a stack of layers, and the headline is only the top one. Before you budget for agents, it helps to understand three things: the ways Agentforce meters usage, the rule that decides which meter you are stuck with, and the costs that never appear on the pricing page but tend to dominate the bill.

This guide walks through all four. The figures below reflect Salesforce's publicly listed rates in 2026, which Salesforce notes are subject to change, so confirm current numbers on Salesforce's official Agentforce pricing page.

The three ways Agentforce is priced

The most important commercial fact about Agentforce is that it is priced on consumption, not on seats. You are largely paying for what the agent does, not for the right to have it available. Salesforce currently offers three live models, Flex Credits, Conversations, and per-user editions, plus a free starting tier. Running three models at once is unusual, and it reflects how hard it is to turn agentic AI usage into a bill a customer can predict in advance.

Flex Credits: pay per action

Flex Credits are Salesforce's consumption currency, introduced in May 2025 and, as of early 2026, the recommended model for most new deployments. Credits are sold at roughly $500 per 100,000. A standard agent action draws about 20 credits, or roughly $0.10, and a voice action draws about 30 credits, roughly $0.15. Usage is tracked in the Digital Wallet, and credits can be bought pre-paid, pay-as-you-go, or on a committed basis.

The detail that surprises teams is that credits are metered per action, not per token, and the credit pool is fungible. The same credits that power agent actions also fund Data Cloud operations and prompts to a bring-your-own large language model. That means the pool drains from more directions than "the agent answered a question," which makes forecasting harder than the per-action rate alone suggests.

Conversations: pay per session

The Conversations model is a flat rate of about $2 per conversation, where a conversation is defined as a coherent session between a person and an agent, effectively a 24-hour window. It is aimed at customer-facing agents and is sold on a pre-purchase basis. Its appeal is predictability, because a fixed rate per interaction is easy to model. The complication is that a single conversation usually triggers several backend actions, so what counts as one billable conversation is not always obvious when one customer question sets off multiple processes.

Per-user editions: the digital-labor option

For employee-facing use, Salesforce also packages Agentforce into per-user editions, framed as "digital labor" pricing. This gives buyers a flat per-user, per-month alternative to usage-based billing. In practice there are several tiers, from an entry-level user license through add-ons and bundled editions built for rolling out employee agents across a company. Two things are worth knowing. Some of these per-user options still require Flex Credits for metered usage, and the bundled editions carry a much higher per-user rate in exchange for a large pool of included credits. Because these tiers move around, treat the per-user path as a category and check the current rates on Salesforce's pricing page rather than a figure you saw quoted elsewhere.

Salesforce Foundations: the free starting point

Salesforce Foundations is a genuine $0 tier, not a time-limited trial. It includes Agent Builder and Prompt Builder along with a starter allotment of Flex Credits and Data Cloud credits, which makes it well suited to a pilot or a limited proof of concept. The catch is that the credit allocation is modest, so real production usage tends to exhaust it, at which point you move to a paid model.

The mutual-exclusivity rule that catches teams out

Here is the rule that quietly shapes the whole decision. Flex Credits and Conversations cannot be used in the same Salesforce org at the same time. You pick one meter, and switching later is expensive to unwind. So the goal of any pricing exercise is to match the meter to how your agents actually behave before you commit.

The rough guide is how chatty your agents are. A deflection bot that resolves a question in two or three actions per session is usually cheaper on Conversations. A multi-step agent that queries records, calls an external system, and updates data burns several actions per session, so Flex Credits is often more honest about what it is really doing, even if the per-session cost ends up higher.

The three models at a glance

The hidden costs that never make the pricing page

The meters above are the top layer. The costs that decide whether Agentforce lands inside budget usually sit underneath, and they are the ones teams underestimate.

Data Cloud is often the biggest line item

Agentforce leans heavily on Data Cloud, also called Data 360, to ground its answers in real customer context. For most deployments this is not optional, and it is frequently the single largest cost in the whole project, often reaching six figures a year. If your Agentforce estimate only counts agent usage, it is almost certainly low, because the data layer that makes agents useful is a separate and substantial spend.

Implementation and services take real work

Standing up agents that behave reliably involves knowledge setup, integration, and professional services. Independent reporting in 2026 has put typical implementation timelines in the range of several months to close to a year, with only a small share of customers reported to have Agentforce fully scaled. None of that appears on a per-action rate, but all of it shows up on the invoice and on the calendar.

Consumption you cannot forecast yet

Because a single conversation consumes several actions, the effective cost of a customer service conversation on Flex Credits usually lands well above the headline per-action rate once you account for the queries, retrievals, and updates a real interaction triggers. Worse, you generally cannot forecast this accurately until you have run agents against real traffic for a couple of months. The honest answer to "what will this cost" is often "we will know after the pilot."

How to estimate your real Agentforce cost

Start from your workflows, not the headline number. Estimate how many times a workflow runs and roughly how many actions each run takes, because that, not the per-action price, is what drives the bill. Model the data layer as its own line rather than folding it into agent usage, since Data Cloud is where the surprises live. Use the free Foundations tier to run a real pilot and capture actual consumption before you commit to a meter, then revisit the Flex Credits versus Conversations choice with data instead of a guess. Above all, resist committing an org to one meter until you have seen how your agents actually behave.

Where a lighter, governed workflow fits

Not every team that prices out Agentforce needs a full agent program on day one. If your immediate goal is a single high-frequency workflow, such as messy intake that has to become clean Salesforce records with a human review before anything is written, a consumption meter and a Data Cloud dependency may be more than the job requires yet.

That is the gap ConvoPro is built for. It is a practical AI workflow layer for Salesforce that turns messy inputs into structured, reviewable Salesforce actions while keeping Salesforce as the system of record, with a review-before-create step and admin control over what the automation can touch. Its pricing model is different from consumption billing: ConvoPro is priced per user per month, with third-party AI and API usage itemized separately, and you can see current plans on ConvoPro pricing. It is designed to complement Agentforce, not replace it, as a way to prove one governed workflow first and decide later whether a broader agent program is justified. If you are weighing options, our guide to Agentforce alternatives covers the wider landscape, and you can always talk to us about a specific Salesforce workflow.

Frequently asked questions

How much does Agentforce cost?

It depends on the model and, more importantly, on the layers beneath it. The headline rates in 2026 are about $2 per conversation or roughly $0.10 per standard action on Flex Credits, but the real total is usually driven by Data Cloud and implementation rather than the agent meter itself. Confirm current figures on Salesforce's Agentforce pricing page.

What is a "conversation" in Agentforce pricing?

Salesforce defines a conversation as a coherent session between a person and an agent, effectively a 24-hour window, billed at about $2 under the Conversations model. The nuance is that one conversation can trigger several backend actions, so a single billable conversation is not the same as a single question.

What are Flex Credits?

Flex Credits are Salesforce's consumption currency for Agentforce, sold at roughly $500 per 100,000 credits. A standard action draws about 20 credits, around $0.10, and a voice action about 30 credits. Credits are fungible, so the same pool also funds Data Cloud operations and bring-your-own-model prompts.

Can I use Flex Credits and Conversations together?

No. The two consumption models are mutually exclusive within a single Salesforce org, and changing your choice later is costly, which is why matching the meter to your agents' behavior up front matters so much.

Is Agentforce priced per user?

There is a per-user, digital-labor option for employee-facing agents, but consumption meters remain the primary models, and some per-user editions still draw Flex Credits for metered usage. Which path costs less depends entirely on volume, so model your own workflows before deciding.

What is the biggest hidden cost in Agentforce?

Data Cloud, or Data 360, is the one budgets most often underestimate. Because agents rely on it for grounding, it is frequently the largest single line item in the project, and an estimate that leaves it out will understate the true cost significantly.