HubSpot Agent Builder, formerly Breeze Studio, lets teams create custom AI agents using their own instructions, knowledge, and CRM data. Agent Hub provides the workspace for managing agents and connecting them to business processes, while the Agent Marketplace offers prebuilt agents that teams can install and customize.

For businesses, the practical opportunity is to turn a repeatable task into a defined process: give an agent a clear job, provide the information it needs, and decide how its work will be reviewed.

This guide explains where to start, when a custom agent makes sense, and what to check before putting one into everyday use.

What happened to Breeze Studio?

HubSpot now identifies Agent Builder as the product formerly known as Breeze Studio. If you previously researched Breeze Studio for custom AI agents, Agent Builder is the current name to look for.

HubSpot describes Agent Builder as a way to build custom agents and workflows on a shared canvas, with Breeze Assistant helping teams create them. Agents can work on individual records or participate in repeatable workflows.

For a broader explanation of the naming changes, read Agent Hub vs. Breeze: What’s the Difference?.

Agent Hub, Agent Builder, and Agent Marketplace: What is the difference?

These tools serve different purposes:

Agent Hub’s Context tab holds business information such as messaging, tone, and ideal customer profiles. Knowledge vaults provide more specific information for targeted use cases. Agentic workflows require a Professional or Enterprise subscription.

The distinction helps you choose your starting point. Browse existing agents when your need is common. Consider building a custom agent when your process requires a more specific approach.

Should you install an agent or build your own?

Start by checking whether an available agent already fits the job.

HubSpot’s Agent Marketplace lists its prebuilt agents. After installation, teams can adjust their instructions, context, and configuration. The marketplace currently does not accept custom agents created in Agent Builder.

Before installing an agent, review its listing for capabilities, default tools, and credit usage. Some HubSpot-built agents require Professional or Enterprise subscriptions.

Our recommendation: write down the outcome you need before comparing options.

“Help our sales team” is too broad to evaluate. “Prepare an internal handoff brief after a deal closes” gives you something concrete to assess.

Choose a prebuilt option if it can meet that requirement with reasonable configuration. Build a custom agent if important parts of your process remain uncovered.

How do you create a custom agent?

In HubSpot, navigate to Agents → Agent Hub, then select Create → Agents.

The builder separates configuration into four parts:

• Instructions: The agent’s role, goal, approach, and expected output.
• Actions: The tools and tasks it can perform.
• Knowledge: Supporting information available across runs.
• Inputs: Information supplied for a particular run.

Creating and customizing agents requires Super Admin or Agent Builder permission. Additional permissions may apply to the features an agent uses. Available actions can include reading or writing CRM records, browsing the web, and connecting to supported external systems through MCP.

Treat these choices as process design decisions. Each action should have a reason to be included, and each information source should help the agent complete its assigned job.

Example: Design an internal sales-to-service handoff agent

Consider a team whose onboarding manager regularly pieces together information after a deal closes.

A possible custom-agent brief could be:

Job: Prepare an internal onboarding handoff for the customer success team.

Expected output: A concise summary of the customer’s goals, purchased services, commitments, and unanswered questions.

Review rule: A team member checks the brief before using it to plan onboarding.

This is a proposed implementation pattern. Its feasibility depends on the information and supported actions available in your account.

Before building it, agree on what a good handoff looks like. Ask the receiving team which details they need and which omissions cause delays.

Then test whether the agent can produce a useful brief from real examples. Include a deal with incomplete notes, one with conflicting details, and one with several associated contacts. Those cases reveal whether the process needs better data, clearer instructions, or more human review.

For preparation guidance, read Do I Need Clean CRM Data Before Using Agent Hub?.

How should you test and control an agent?

HubSpot documents controls for assigning agent access, testing outputs, reviewing estimated credit costs, and setting run limits. Builder tests can run without consuming HubSpot Credits; actual production costs can differ from estimates.

We recommend testing against written acceptance criteria. For the handoff example, those might include:

• Customer commitments are represented accurately.
• Missing information is clearly identified.
• The brief follows the agreed structure.
• The reviewer can trace important statements to supporting information.
• Preparing and reviewing the brief takes less time than the current process.

Keep responsibility clear. Name the person who maintains the instructions, the person who reviews results, and the person who decides whether the agent is ready for broader use.

For related guidance, read Agent Hub Security: What Can It Read?.

Does Agent Builder cost extra?

HubSpot states that Agent Builder is included in Starter, Professional, and Enterprise editions. Running custom agents consumes HubSpot Credits, with work measured in action units. Credit consumption varies with the actions performed during a run.

Budget around a defined task and expected volume. A useful pilot should measure both the effort saved and the cost of producing an acceptable result.

For a broader explanation, read HubSpot Breeze Pricing.

Your first custom-agent checklist

Before expanding beyond a pilot:

• Choose one repeatable task.
• Check whether a prebuilt agent meets the need.
• Define the expected output.
• Identify the necessary information and actions.
• Assign an owner and reviewer.
• Test ordinary cases and difficult cases.
• Measure accuracy, review effort, and credit usage.
• Decide what needs to improve before wider adoption.

An agent belongs in a process when its results make that process measurably better. For some tasks, that means faster preparation. For others, it means more consistent information or fewer missed steps.

Use How Do I Know If Agent Hub Is Working? Measuring ROI to connect the pilot to a business outcome.

Build custom agents around work your team understands

The strongest starting point is a task with a clear owner, usable information, and a result people can evaluate.

Fast Slow Motion can help you identify those tasks, assess your HubSpot setup, and design a practical implementation approach.

Explore Fast Slow Motion’s HubSpot services.

Sources

• HubSpot Agent Builder product overview
• Understand Agent Hub
• Create and customize agents in the agent builder
• Install agents from the Agent Marketplace