HubSpot Agent Hub, Agent Builder, and Agent Inbox split the work of running AI agents into three jobs. Agent Builder is where you create custom agents and give them instructions, actions, and data. Agent Hub is the home base where you can activate, monitor, and manage every agent. Agent Inbox is where you can run agents and review, approve, refine, or share what they produce.
Teams start with a few agents, add custom ones over time, and eventually struggle to keep track of what each one does. Knowing where each job lives keeps your agents working from the same picture of the customer, and it makes credit costs easier to control.
If you're planning your first agents and want them built on clean data, our HubSpot AI agent implementation team can help you plan, build, and govern them.
HubSpot Agent Hub is the central place to see and manage every AI agent running across marketing, sales, and service. From one screen, you can check live status and performance for each active agent, turn on inactive agents with one click, and open Agent Builder or the Agent Marketplace.
Agent Hub also groups agent results by go-to-market goal. The four goals are building demand, winning deals, delighting customers, and scaling growth. You can open it from Agents > Agent Hub in your portal.
Agent Hub has three tabs, called Intro, Agents, and Context. The Intro tab shows HubSpot's pre-built agents, including the AEO, Data, Prospecting, Deal Progression, and Customer agents. The Context tab holds shared business information, such as brand, tone, ICPs, and knowledge vaults.
The problem Agent Hub solves is that agents don't know about each other. HubSpot's own example is a prospecting agent emailing a customer the same week a service agent is handling that account's open complaint. When every agent shows up in one place, someone can catch that overlap before the customer does.
It’s best to turn on one pre-built agent first and let it run for a few weeks before you build your own. You'll learn how agents behave on your data, and Agent Hub will give you a baseline to compare custom agents against.
Agent Builder lets you create custom agents and automations that run on the data already in HubSpot. You can describe the job in plain language through Breeze Assistant and let Agent Builder work out what to run, or you can set up each part of the agent yourself.
Every custom agent is built from four settings in the agent builder.
|
Part |
What it controls |
Example |
|
Instructions |
The agent's role, goal, approach, and expected output |
Summarize a closed deal for the onboarding team in three sections |
|
Actions |
What the agent can do, from HubSpot actions to web browsing, reading or writing CRM records, and MCP connections to outside systems |
Read the deal, company, and recent activity |
|
Knowledge |
Context the agent reuses on every run, such as knowledge vaults, your brand kit, ICPs, or integration data |
Your onboarding playbook stored in a knowledge vault |
|
Inputs |
The data for one run, such as a CRM record or a value a user types |
The deal that just closed |
Agents can start from almost any trigger, including a schedule, a contact update, a webhook, or a third-party integration. You can place workflows, custom agents, and triggers on one canvas, so an agent becomes a step inside a larger process.
Agent Builder can only use HubSpot's actions, web browsing, and the outside systems you can reach through MCP. Since September 2026, developers can add an MCP server to a HubSpot app, and its tools show up in Agent Builder automatically.
When an agent needs a tool with no MCP server or an action HubSpot doesn't offer, that's a job for custom AI agent development. For tasks that follow fixed rules, a standard HubSpot workflow is still cheaper and easier to audit than an agent.
Agent Inbox is where you can run agents, review what they produce, and act on it. To run an agent, go to Agents > Agent Inbox, pick it from the Run agents menu, fill in its inputs, and click Run agent.
Every run in the inbox carries one of five statuses, and you can filter by each one.
|
Status |
What it means |
|
Needs approval |
The run is waiting for a person to approve a tool or action |
|
Running |
The agent is working on it now |
|
Waiting |
The run is queued or paused before it starts |
|
Success |
The run finished without errors |
|
Failed |
The run hit an error and stopped |
Each run shows the thinking steps behind it, the sources the agent used, the user who started it, and the estimated credits it cost. From there, you can take four actions.
During the first month, check the Needs approval filter daily. Agents won't complete tasks that require approval until someone reviews them. Checking this queue regularly helps you catch pending approvals before they hold up the work.
An agent moves through five steps, from build to review, and each step lives in one of the three tools. Here's the order we follow on client portals.
Agent Hub launched for Professional and Enterprise customers, and the agent builder now also lists Starter tiers of Marketing, Sales, Service, Data, and Content Hub.
Smart CRM, Revenue Hub, agentic workflows, and some HubSpot-built agents still require Professional or Enterprise. Custom agent runs use HubSpot Credits.
|
Tier |
Where you can run a custom agent |
|
Starter |
The automation side panel, a CRM card, Agent Inbox, or Breeze Assistant |
|
Professional and Enterprise |
A workflow in the Agent Hub workflow builder, a CRM card, Agent Inbox, or Breeze Assistant |
You need Super Admin or Agent Builder permission to create and edit agents. By default, Super Admins can edit agents and every user can run them, but you can limit both to specific users and teams. When an automation triggers an agent, the agent runs with the agent owner's permissions, so an owner without access to contacts gets an agent that can't read contacts.
If your user roles have grown messy over the years, our HubSpot User Permissions Guide walks through cleaning them up before agents inherit them.
A custom agent uses HubSpot Credits when it does work, such as thinking, taking actions, and generating content. Costs change with how complex the task is, so test the agent first and review its estimated credits per run. If an agent is added to a deal-closed workflow and 50 deals close in a month, that's 50 paid runs. Set a monthly run limit on every agent before you add it to a workflow.
Use a pre-built agent when HubSpot already ships one for the job, and build a custom agent only when the process is specific to your business. Pre-built agents need less setup, and custom agents take testing and upkeep.
|
Question |
Pre-built agent |
Custom agent |
|
Does HubSpot ship an agent for this job? |
Yes |
No |
|
How much setup does it take? |
Activation and a few settings |
Instructions, actions, knowledge, inputs, and testing |
|
Who keeps it working? |
HubSpot built it, and your team adjusts its settings |
Your team owns the instructions |
|
Best fit |
Common jobs like customer support or prospecting |
A step only your business does, like Ignite Reading's calendar lookup |
These are the rules we follow when clients ask us to build their first custom agents.
A custom agent that drafts and a person who approves is a better first build than an agent that updates records on its own.
Set up your AI settings, clean the data agents will read, tighten permissions, and name an owner for every agent before you activate anything. Skipped setup is the most common reason we see first agents produce bad output.
A portal that passes these five checks is ready for its first custom agent.
Most agent problems we fix start before the first build, with messy data, broad permissions, and nobody assigned to review output. Our team audits your portal, sets up AI settings, permissions, and run limits, and builds and tests custom agents in Agent Builder before they touch live records.
As a HubSpot Elite Solutions Partner, we plan agents around the way your sales, marketing, and service teams already work. If you're ready to put agents to work, learn how our HubSpot AI agent implementation service works.
Recap
Agent Builder creates custom agents, Agent Inbox runs them and holds their output for review, and Agent Hub keeps every agent visible in one place. Start with one pre-built agent, test custom agents before they spend credits, and keep write access off until the output earns your team's trust.
Before you build your first custom agent, check how ready your portal is with the free HubSpot AI Readiness Scorecard.