HubSpot Credits: What AI Agents Cost and How to Cap Them

HubSpot Credits are how HubSpot bills for work its AI agents and some automation features do. HubSpot's help center says that on July 23, 2026, all of its AI agents started using credits, and it recommends testing each agent before you turn it on so the bill does not surprise you. A credit costs $0.010. The Customer Agent, the one most small teams try first, uses 50 credits ($0.50) for each conversation it resolves.

This post covers what each agent costs, what HubSpot counts as a resolution, what happens when you run out, and which limits to set before an agent goes live.

We read these HubSpot pages on October 19, 2026: the HubSpot Credits product page, the credits section and rate sheet in HubSpot's Products and Services Catalog, the help articles on managing HubSpot Credits, testing and capping agent credit use and the Customer Agent, and the Customer Agent product page. Every HubSpot fact below comes from those pages. We have not run a billing cycle on a test account for this post.

Paid HubSpot plans include a monthly pool of credits. For most hubs, Starter plans get 500, Professional plans get 3,000 and Enterprise plans get 5,000. The amount follows your highest tier and is not added up across hubs. Credits reset each month and unused ones expire. Extra credits come in packs of 1,000 for $10, or you can pay per credit past your limit. Agents spend credits per action, at rates set in HubSpot's rate sheet. Limits on spending are not turned on by default, so set them yourself.

These are the rates in HubSpot's standard rate sheet. HubSpot notes that accounts on its newer flexible seats-and-credits pricing have a separate rate sheet with different rates, so check which one applies to you.

• Customer Agent: 50 credits per resolved text conversation, and 50 credits per minute of voice conversation (voice is in beta)
• Prospecting Agent: 100 credits per lead it recommends for outreach
• Data Agent: 10 credits per prompt, per record
• Nurture Agent (beta): 10 credits per personalized email
• Knowledge Base Agent (beta): 200 credits per new article
• Revenue Agent (beta): 500 credits per invoice it starts collecting on
• Content Agent (beta): 1,000 credits per piece of content
• Custom agents (beta): 1 credit per action unit, with the estimate shown when you test the agent

HubSpot also says beta features may start using credits later and rates can change. It aims to give 30 days' notice before a beta feature starts charging, but says that may not always be possible.

You pay for resolutions, not messages. A conversation counts as resolved in one of two ways. The first is when the agent replied at least once with a content source (such as a help article) or took an action (such as a password reset), and the visitor did not ask for a person within 72 hours of their last message. The second is when the agent marks a lead as qualified, partially qualified or not qualified.

Three details change the math:

1. 1Only a handoff the visitor starts stops the charge. If your own workflow assigns the conversation to a rep, it can still count as resolved.
2. 2The status is set 72 hours after the visitor's last message. A rep stepping in after that, or bad feedback, does not reverse it.
3. 3If a customer replies to an email thread more than 72 hours later, the conversation reopens. A second resolution costs another 50 credits.

Reply suggestions inside the help desk, where a rep reviews and sends the answer, do not use credits. That is a useful first step for a team that wants to see the agent's answers before customers do.

If you only have the included credits, features that use credits pause until your monthly reset or until you buy more. For the Customer Agent, that means it stops picking up new conversations on every connected channel, and new chats follow your handoff setup instead.

If you buy extra credits, the rules change. Going past your monthly amount triggers one of two settings. Auto-upgrade moves you to a bigger credit pack for the rest of your contract. Pay-as-you-go bills the extra credits in arrears, in steps of 10, and your limit goes back to normal at the next reset. HubSpot made pay-as-you-go the default for packs bought on or after September 16, 2026; packs bought before that default to auto-upgrade. Packs can only be reduced at the end of your contract term, so an auto-upgrade can stick for months.

HubSpot gives you four controls. We would use all of them.

1. 1Test runs. Each agent has a Test button in Agent Hub that runs it without using credits and shows the input, reasoning and output without changing CRM records. HubSpot warns that a test can still call any MCP tools the agent is set up with, so use sample data. Test history shows the estimated credits per run.
2. 2A monthly run limit per agent. Set in the agent builder under Edit run limit, it stops that agent once it has run the number of times you chose, and shows an estimated credit range as you type.
3. 3A feature limit. Under Account and Billing, HubSpot Credits, you can cap how many credits one tool may use in a month. It caps that tool and does not set credits aside for it.
4. 4An account limit. After you buy extra credits, HubSpot sets an account cap at 50% above your monthly total and emails Super Admins and Billing Admins at 75%, 85% and 90% and when the cap is hit. You can lower it.

You can also pause Customer Agent credit use from its settings at any time. HubSpot says it can take a few hours for usage to fully stop, so pause a few hours before you change content sources or the agent's setup.

This is arithmetic from the published rates, not a client result. A Professional account has 3,000 included credits. If nothing else spends them, that covers 60 Customer Agent resolutions a month. A team expecting 300 resolutions would need 15,000 credits. That is 12,000 more than the included pool, which at $0.010 each is $120 of extra credits that month. Add a Prospecting Agent recommending 100 leads and that is another 10,000 credits. Run the numbers for each agent you plan to use before choosing a pack.

These are designs we would set up. None of them describes a client result.

1. 1A support agent with a budget. Turn on reply suggestions first and let reps send the answers for two weeks. Then move the Customer Agent to chat for the questions your help articles already answer well, with a run limit sized to your expected volume and a feature limit just above it. Review the resolution list each week to see whether paid resolutions match questions that were actually answered.
2. 2A spend report outside HubSpot. A weekly job reads credit use per feature and posts it to the team's chat with the month-to-date total and the share of the cap used. HubSpot's own alerts start at 75%, which can be late for a small budget.
3. 3Agent or workflow. For high-volume, rule-based jobs, such as tagging leads by form answer or routing tickets by keyword, a plain HubSpot workflow or a small custom integration often costs nothing per run. We keep agents for the steps that need judgment. Our post on AI agents vs. automations explains how we draw that line.

For any of these we would write down what each agent may change, test it on sample records, and set a cap per agent, the same way we size agent action budgets. Our pre-production testing checklist covers the test cases.

The Customer Agent needs a Professional or Enterprise plan. If you are on Starter and the agent is the only reason to upgrade, compare that cost with a custom support agent or a no-code tool first. Our post on Zapier's AI step and Zapier Agents covers one cheaper route for simple jobs. If your help articles are thin, the agent will hand off often or answer badly, and fixing the articles comes first. If a job runs thousands of times a month and follows fixed rules, a workflow is cheaper and easier to check. And if your contract defaults to auto-upgrade, switch to pay-as-you-go or set an account cap before testing at scale, since a bigger pack cannot be reduced until renewal.

OrchestriAI is not a HubSpot partner or reseller. We connect HubSpot to the rest of a business through its APIs, as described on our HubSpot integration page, and we build AI agent systems and workflow automation with clear limits on what they can spend and change. If you want to know which HubSpot jobs are worth an agent and which are cheaper as a workflow, start with the free automation audit.

OrchestriAI is an independent implementation provider. Product names are used for identification only. OrchestriAI is not affiliated with, endorsed by, or sponsored by HubSpot. This article summarizes HubSpot's public pricing pages and help articles as of October 19, 2026. Rates, included credits and settings can change, and accounts on flexible seats-and-credits pricing use different rates, so confirm them in your own account and HubSpot's current rate sheet. This is not financial or legal advice.