HubSpot Breeze vs custom AI agents: which should you use?

Most HubSpot customers meet AI through one name, Breeze, and then discover that the name covers several quite different things. HubSpot's own documentation describes an Agent Hub that lists a customer agent, a prospecting agent, a data agent, a deal progression agent and an AEO agent, each with its own subscription and credit requirements. That is useful if your work lives inside HubSpot. It is less clear if your workflow crosses billing, email, spreadsheets and a field service tool.

This guide compares HubSpot Breeze agents with custom AI agents built around your own processes. It is written for owners and operations leads at small and mid-sized firms who already pay for HubSpot and want to know where the built-in agents are enough, where they stop, and what a custom build would add. We cover what each Breeze agent does, how the credits model works, the limits to plan around, and a practical way to decide.

Quick answer

Use HubSpot Breeze agents when the job sits inside HubSpot, such as answering support questions from your knowledge base, researching leads or cleaning CRM records. Choose a custom AI agent when the work spans several systems, needs your own business rules, or must run outside HubSpot's credit model. Many firms sensibly use both, with the custom agent writing back to the CRM.

What HubSpot Breeze agents actually are

Breeze is HubSpot's umbrella brand for its AI features. In practice the part that matters for automation is Agent Hub. HubSpot's documentation says Agent Hub lets you view all agents and agentic workflows running across your account, and lists five agents in its customer journey: an AEO agent that improves brand visibility across AI-powered search, a data agent that enriches and maintains CRM data, a prospecting agent that researches target companies and drafts personalised outreach, a deal progression agent that recommends next steps to move deals along, and a customer agent that responds to support conversations. The same page also notes that some HubSpot-built agents need a Professional or Enterprise subscription, and that agentic workflows need Professional or Enterprise too.

Two things follow from that list. First, each agent is designed around a job HubSpot already understands: contacts, companies, deals and conversations. Second, every one of them reads and writes HubSpot data. That is a strength when your process is a clean CRM process and a limitation when it is not.

It also helps to separate three layers that the marketing tends to blur. There are the packaged agents that HubSpot ships and maintains. There are workflow actions that call an agent as one step in an automation. And there is the developer platform, where you can add your own tools that an agent may call. A custom agent in the sense used in this article is a fourth thing again: a purpose-built agent that sits outside HubSpot, reads from several systems, and writes back where needed.

What each Breeze agent is good for

Customer agent

The customer agent is the most mature of the group for most small firms. HubSpot says it answers customer questions using your existing content, such as synced knowledge base articles and files, and can perform configured actions like resetting a password or booking a meeting. It can run on live chat, Facebook or WhatsApp, and it hands off to a human when its confidence is low or escalation rules are met. There is also a lower-risk mode where it only suggests replies to your reps inside the help desk, which HubSpot says uses no credits.

The limits are worth reading closely. HubSpot states that if the agent does not know an answer it asks the visitor to rephrase or transfers to a human, and that chats with no visitor reply for 24 hours close automatically. It needs a Professional or Enterprise subscription in a qualifying hub such as Service Hub, plus an assigned seat. Its answers come from the content you have synced, so the quality of the agent is capped by the quality of your knowledge base.

Prospecting agent

The prospecting agent researches target companies and drafts personalised outreach. It suits sales teams that already keep clean company and contact data in HubSpot and want a first draft of research and messaging for each lead. It is less suited to firms whose best signals live elsewhere, for example in a tender portal, an industry directory or an inbox, because the agent works from what it can reach.

Data agent

The data agent enriches and maintains CRM records. For firms with messy contact data this is often the quickest win, since duplicate, incomplete or stale records undermine every other agent. It does not fix a process problem, though. If the same customer is spelled three ways across HubSpot, your accounting package and your project tool, tidying one system leaves the others untouched.

Deal progression and AEO agents

The deal progression agent gives sales teams recommendations to move deals through the pipeline, while the AEO agent targets visibility in AI-powered search and requires a Marketing Hub Professional or Enterprise subscription, according to HubSpot's documentation. Both are useful inside their lane. Neither is a general-purpose worker you can point at an arbitrary task.

How the credits model works

Breeze agents are not simply included with your subscription. HubSpot's documentation says usage-based features such as the customer agent, prospecting agent and data agent consume HubSpot Credits, as do specific actions like running a Breeze action in a workflow. All accounts on seat-based pricing receive included credits based on their highest subscription tier. Credits reset monthly and unused credits expire rather than rolling over. Only users with paid seats can use them.

The rate sheet in HubSpot's product catalogue gives concrete numbers. HubSpot lists 50 credits for one resolved conversation by the customer agent, 100 credits for one prospecting outreach recommendation, 10 credits for one data agent response on one record, and 10 credits for one AI action in a workflow. The same page warns that some actions may also incur separate charges such as telephony minutes or SMS messages.

For overage, HubSpot's billing page lists a rate of $0.010 per credit, billed monthly in arrears in increments of 10 credits. At that rate, 50 credits for a resolved conversation is 50 cents and 100 credits for a prospecting recommendation is one dollar. Those are small numbers per task, and for a modest volume of well-defined work they can be very good value. The point is to model them against your own volumes rather than assume they are free.

Two behaviours deserve a place in your planning. First, if you exceed your limit with no purchased packs, usage-based features pause until the next reset. For a customer agent on live chat, HubSpot says new conversations then follow your handoff process, which means your team absorbs the overflow. Second, the same page says packs bought on or after 16 September 2026 default to pay-as-you-go overage, whereas earlier packs default to automatic upgrade. If you bought capacity earlier in the year, check which setting applies to you before a busy month.

A final detail: how credits are counted for the customer agent is not by message. HubSpot says credits are consumed per resolution, defined as a reply that shares a content source or performs an action, with no qualifying handoff within 72 hours of the visitor's last response. That is a sensible definition, but it means the cost you see this week lags the work done, because resolution status is evaluated 72 hours after the last visitor message.

Where Breeze agents stop

None of the following is a criticism of the product. They are the places where a packaged agent will meet the shape of a real small business.

The first limit is data reach. HubSpot's developer documentation for agent tools is blunt on this point: during beta, the agent does not have access to CRM data unless explicitly given tools to get that data. It also says the agent cannot ask the user for more information after it has been invoked, so it must receive enough input to finish the task. Those are sensible design choices, but they mean multi-step work that needs clarification mid-task is awkward.

The second limit is how far a workflow can run an agent. HubSpot's workflow documentation says the Run Agent action is in beta and is limited to 500 executions per day, and that you must install agents from the marketplace before they appear. It also requires a Professional or Enterprise subscription at the relevant hub and HubSpot Credits. For a firm processing a few hundred items a day that ceiling may never matter. For a firm that batches thousands of records, it will.

The third limit is process shape. Packaged agents are built for the process HubSpot expects: a lead becomes a deal, a deal closes, a customer asks for help. Many small firms run something different. A recruiter matches candidates to roles across an applicant tracking system, a job board and an inbox. A property manager chases maintenance across a tenant portal and contractor emails. An accountancy practice reconciles client documents against a ledger. HubSpot may hold part of that picture, but not all of it.

The fourth limit is control. You can configure what a Breeze agent may see through AI settings, and you can write handoff rules, but the model, the retrieval and the orchestration are HubSpot's. If a regulated client asks exactly how an answer was produced, or you need a specific approval step before an action runs, you are working within the options HubSpot exposes.

Extending Breeze with your own tools

There is a middle path that many firms overlook. HubSpot's developer platform lets you add custom agent tools, so your own logic can sit behind a HubSpot agent.

According to the developer reference, agent tools are custom workflow actions that HubSpot AI agents can invoke to complete tasks in a chat. A tool is declared as one of three types: GET_DATA to retrieve information, GENERATE to create content or suggestions, or TAKE_ACTION to perform actions such as creating notes or tasks in HubSpot or in external systems. A tool sends input fields to an external endpoint and returns output fields, and one tool can be used across several agents and in workflows.

The constraints are specific. The action URL must be a publicly accessible endpoint rather than a serverless function inside the developer project, output fields must be string key-value pairs, and the documentation warns that non-string values cause the outputs to be ignored. Requests arrive as POST calls with a signature header for validation, and blocked executions expire after one week by default. You should not pass secrets through input fields.

This approach is attractive when the front door should stay HubSpot, for example a customer agent on your website, but the answer needs data from your billing system or a bespoke calculation. It does mean you now maintain an endpoint, and HubSpot's credit model still applies to the agent that calls it.

API limits if you build outside HubSpot

When a custom agent reads and writes HubSpot from the outside, the API usage guidelines apply. HubSpot's documentation lists, for privately distributed apps, a burst limit of 100 requests per 10 seconds per app on Free and Starter, and 190 on Professional and Enterprise, with daily limits per account of 250,000 on Free and Starter, 625,000 on Professional and 1,000,000 on Enterprise. The daily limit is shared by every app in the account, and exceeding a limit returns a 429 response.

For most small businesses these ceilings are generous. They matter when an agent does a lot of searching, since the same page says the CRM Search, Associations and GraphQL APIs have stricter limits than the general ones. A well-built agent batches requests, caches what it can and backs off on a 429, which is ordinary engineering rather than a reason to avoid the approach.

How to decide: a practical test

Instead of choosing by brand, run the work through five questions.

Our wider comparison, custom AI agent vs off-the-shelf tool, goes through the same trade-off for any platform rather than HubSpot alone. If you are weighing how a new agent would connect to the tools you already run, how AI agents integrate with existing SMB tools covers the practical patterns.

The risk of buying the label

It is sensible to be sceptical of any vendor's agent claims, including ours. Gartner has described a trend it calls agent washing, where vendors relabel existing chatbots, robotic process automation and AI assistants as agentic AI without meaningful agentic capability. Gartner estimated that only about 130 of the thousands of vendors claiming agentic AI offerings are real, and predicted that over 40% of agentic AI projects will be cancelled by the end of 2027, citing escalating costs, unclear business value or inadequate risk controls.

The lesson for a HubSpot customer is not that Breeze is a label. Its documentation is specific about what each agent does and what it costs in credits. The lesson is to write down the job, the volume and the measure of success before you switch anything on, and to review the credit usage after the first month. The same discipline applies if you choose custom work. Our article on what can go wrong with AI agents in production lists the failure modes worth testing for.

How Klevere approaches HubSpot and AI agents

We do not treat this as a choice between HubSpot and everything else. Klevere has deployed 500+ AI agents across 50+ projects and 12 industries, and the pattern we see most often is a hybrid. HubSpot stays the system of record and the place your team works. Breeze agents handle the jobs that sit squarely inside it. A custom agent handles the work that crosses systems or needs rules specific to your firm, and writes its results back to HubSpot so nobody has to look in two places.

Our process starts with an audit of the workflow, not the tool. We map each step, mark where data comes from and where it must end up, and estimate the volume. Only then do we say whether a packaged agent, a custom tool behind a HubSpot agent, or a standalone agent fits best. Sometimes the honest answer is that Breeze is enough and a custom build would be wasted spend.

When a custom build is the right call, it is scoped with human approval points where an action is hard to reverse, logging of what the agent did and why, and a handover so your team can run it. If you want to see how we connect to the CRM itself, our HubSpot AI integration page describes the approach, and our custom AI agent development service covers the build.

Frequently asked questions

What is HubSpot Breeze?

Breeze is HubSpot's name for its AI features. In day-to-day use it mostly means the agents in Agent Hub, such as the customer, prospecting and data agents, plus AI actions in workflows. Some need a Professional or Enterprise subscription, and the usage-based ones consume HubSpot Credits rather than being included at no extra cost.

Are HubSpot Breeze agents free?

Not entirely. Accounts on seat-based pricing get included credits tied to their highest tier, and these reset monthly without rollover. Usage-based agents consume them. HubSpot's rate sheet lists, for example, 50 credits per resolved customer agent conversation. Once your allowance is used, features pause or overage applies, depending on your settings.

Can HubSpot Breeze agents connect to other software?

Yes, through custom agent tools built on the developer platform. A tool can send fields to a public endpoint you control and return string outputs, so an agent can fetch data from or act in an external system. You must build and maintain that endpoint, and HubSpot notes the agent has no CRM access unless given tools for it.

When is a custom AI agent better than Breeze?

A custom agent tends to win when the task spans several systems, needs your own business rules, requires a defined approval step, or runs at volumes that make credit-based pricing awkward. If the job lives entirely in HubSpot records and conversations, a Breeze agent is usually quicker to start and cheaper to maintain.

Can I use both Breeze and a custom agent together?

Yes, and many firms should. Keep HubSpot as the system of record, let Breeze handle the in-CRM jobs, and have a custom agent handle cross-system work and write results back through the API. The main thing to design carefully is ownership, so two agents never edit the same field for different reasons.

Next step

If you are unsure which side of the line your workflows fall on, the quickest way to find out is to map them. Book a free AI audit and we will tell you plainly where Breeze is enough and where a custom agent would pay for itself.