Agentforce pricing explained: credits, conversations and add-ons
Ayo Adeniran · Oct 28, 2026 · 5 min read
Agentforce is Salesforce's platform for AI agents, and its pricing page offers several ways to pay: per action, per conversation, per user, or bundled into an edition. Each model rewards a different kind of usage, and picking the wrong one can multiply the bill.
This post explains each model, works through what they cost at realistic volumes, and covers when Agentforce is worth it and when another approach fits better.
What Agentforce is
Salesforce describes Agentforce as "the AI agent platform that delivers 24/7 autonomous support at enterprise scale." Its agents answer from "trusted business data, including Salesforce CRM data, external data from Data 360, and more." You build them in the low-code Agent Builder or with Agent Script, connect other systems through MuleSoft API connectors, and use MCP servers from AgentExchange partners.
Its strength is obvious: if your customers, cases and opportunities already live in Salesforce, the agent starts next to the data.
The pricing models
| Model | Price | How it works |
|---|---|---|
| Flex Credits | $500 per 100,000 credits | An Agentforce action uses 20 credits, $0.10. A voice action uses 30, $0.15 |
| Per conversation | $2 per conversation | Bought in advance only |
| Agentforce add-ons | $125 per user a month | For Sales, Service and Field Service; $150 for Industries clouds |
| Agentforce User License | $5 per user a month | Requires Flex Credits |
| Agentforce Coworker User License | $50 per user a month | |
| Max Edition | From $550 per user a month | Includes the Agentforce add-on and 2.75 million Flex Credits a year |
| Salesforce Foundations | $0 | To get started with Agentforce Builder, Prompt Builder and Agent Script |
Flex Credits can be bought up front, committed in advance or paid as you go. Per-conversation pricing is pre-purchase only.
What it costs in practice
The figures below are worked from the published prices above. They are illustrations, not quotes; your contract and discounts will differ.
Actions or conversations? The two consumption models cross at 20 actions per conversation, because $2 buys 20 actions at $0.10 each.
| A customer service agent | Flex Credits | Per conversation |
|---|---|---|
| 5,000 conversations a month, 4 actions each | 20,000 actions = $2,000 | $10,000 |
| 5,000 conversations a month, 25 actions each | 125,000 actions = $12,500 | $10,000 |
| The same agent on voice, 4 voice actions each | 20,000 voice actions = $3,000 | $10,000 |
Short, focused conversations favour Flex Credits. Long, tool-heavy conversations can tip the balance toward per-conversation pricing, but only if you can forecast volume well enough to buy it up front.
Per-user add-ons suit employee-facing agents with a known headcount. Fifty service reps on the $125 add-on is $6,250 a month, whatever they ask.
The quiet cost is the action count. An agent that looks up an order, checks a policy, drafts a reply and updates the case may take several actions per turn. Measure that number in a pilot before choosing a model, because every price above scales with it.
Is Agentforce worth it?
It is a strong fit when:
- The work starts and ends in Salesforce: service cases, sales follow-up, field service.
- Your data is already in Salesforce CRM and Data 360, so grounding needs no new pipeline.
- You want customer-facing channels: Agentforce extends agents to the web, the phone and apps.
It fits less well when:
- The process crosses many systems outside Salesforce, and most of the work happens elsewhere.
- Steps wait days for a person in finance, legal or operations, and the agent must keep its place.
- You want to choose models per step, or run calls on your own provider contracts.
The alternatives
Microsoft Copilot Studio is the equivalent path for organizations on Microsoft 365, covered in Microsoft Copilot Studio alternatives.
n8n, Zapier and Make connect thousands of apps and now run AI agents, each with its own way of counting usage. See n8n vs Zapier vs Make for AI agents.
Momor Workflows is for the processes that cross systems, wait on people and must show afterwards what happened.
Where Momor fits
Momor Workflows runs a process across AI, software and people, and writes every step down as it happens.
- Systems through MCP. Connect an MCP server and its tools become steps, signed in as the organization or as each member. A tool can be marked as needing approval, so the call waits for the run's owner to allow it.
- Waits that last. A Wait step asks a person and the run keeps its place for hours or days, holding no worker. After a restart, the run continues from the first step that did not finish.
- Models per step. Every model call goes through Momor Gateway, so each step can use a different model, from a different provider, with failover when one fails. Calls can run on your own provider keys.
- Assistants as the front door. Your chosen workflows can be offered as MCP tools to claude.ai, ChatGPT and other clients, where people sign in as themselves.
- Priced with you. Workflows is sold under an agreement, around the workflows you run, how often they run and the models they call, with unlimited workflows, steps and members.
What to weigh: Momor is not a customer-service channel product. People reach workflows through Momor's Chat, the API or an MCP client, not through a phone line or a website widget. Runs start from those doors, with no schedule, email or webhook trigger yet. And connecting an MCP server, including one for Salesforce, is done with Momor's help for now.
If your agent lives in Salesforce, start with Agentforce and measure your action count. If your process lives across your systems and your people, read the Workflows quickstart or talk to us.