How to Add AI Agents to HubSpot for Sales and Support: Breeze Agents Step by Step

"Add AI agents to HubSpot" sounds like one project. In practice it is two, because the agent that answers your customers and the agent that finds your next customers are different products with different requirements, different setup screens and different billing events. This is the order we set each one up on client portals, what you need before you touch either, and what a real month costs.

Every HubSpot requirement below was read in HubSpot's own knowledge base on 26 September 2026. Where a figure comes from elsewhere, the source and date are next to it.

What do you need before you add an AI agent to HubSpot?

Three things: a subscription that includes the agent, HubSpot Credits, and the right AI settings and permissions. The Customer Agent needs a Professional or Enterprise tier; the Prospecting Agent works from Starter on most Hubs. Both need AI access switched on in settings, and a clean portal, because an agent acts on whatever record it reads.

The specifics, from HubSpot's knowledge base:

• Customer Agent. Professional or Enterprise on Marketing, Sales, Service, Data, Content or Revenue Hub, or Smart CRM. Users need the Customer agent editor permission and an assigned seat. Source: Set up the customer agent, last updated 10 August 2026, read 26 September 2026.
• Prospecting Agent. Starter, Professional or Enterprise on Marketing, Sales, Service, Data or Content Hub, or Smart CRM Professional and Enterprise. Users must be a Super Admin or hold the Access prospecting agent permission, and the AI settings for generative features, CRM data, customer conversation data and files data must be on. Credit actions do not run in sandbox accounts. Source: Set up and use the prospecting agent, last updated 16 September 2026, read 26 September 2026.

The requirement HubSpot cannot check for you is the data. Duplicate contacts, picklists with three spellings of one industry and fields that disagree with your billing system make an agent confidently wrong. We covered the five portal problems that break agents at go-live separately; the AI readiness checklist scores a portal in fifteen checks.

Which Breeze agent handles sales, and which handles support?

The Customer Agent handles support: it answers customer questions from the knowledge you approve and hands the rest to a person. The Prospecting Agent handles sales: it watches target companies for buying signals, finds contacts and drafts outreach for a rep. The Data Agent is a useful third for sales teams who want answers about records and calls.

Credit rates as published on INSIDEA's agent activation page, transcribed from HubSpot's product and services catalog and verified there on 5 September 2026. HubSpot changes rates; check the credits breakdown before you forecast. If you are still deciding which one to start with, Customer vs Prospecting vs Data compares the three.

How do you set up the Customer Agent for support, step by step?

Create the agent, give it an identity, point it at your knowledge, set what it may do and when it hands off, test it, then deploy it to one channel. HubSpot's own setup sequence has these steps, and testing in preview does not consume credits, so there is no cost to getting the test stage right.

1. Identity. A name, an avatar and a personality. HubSpot offers Friendly, Professional, Casual, Empathetic and Witty. Pick the one your support team already writes in.
2. Knowledge sources. Add your help centre, product documentation and public pages. You can crawl a whole domain or match specific paths; match paths. Crawling marketing pages teaches the agent to answer support questions in sales copy.
3. Permissions. Decide which CRM properties the agent may update. Start with none, or with a single ticket property you want it to set.
4. Actions and guidelines. Actions such as a password reset or lead qualification, and written guidelines for tone, topics to refuse and phrases to avoid.
5. Handoff. When the agent escalates, and to whom. Write it down as a rule: after two failed answers, on any billing dispute, on any cancellation mention.
6. Test. Preview on email and live chat, read the testing insights, and coach the agent on the answers it gets wrong. We run the fifty most common questions from the last quarter's tickets through it before go-live.
7. Deploy. One channel first, usually live chat, then email once the first week's escalations look right.

One thing to know before you create it: HubSpot's documentation says a customer agent cannot be deleted once created, only paused. Name it for the brand or channel it serves.

How do you set up the Prospecting Agent for sales, step by step?

The Prospecting Agent runs on plays: a set of instructions for who to target, what signals to watch, how to write and what it may not do. You build a play, test it against a real contact, then turn it on. Everything useful about this agent lives in how tightly the play is defined.

1. Open it. In HubSpot, go to Sales, then Prospecting Agent, and click Create play.
2. Audience. Choose the company segments to monitor and the intent signals to watch, such as funding news, product launches, research on your topics or visits to your site. Add the personas; the agent finds up to three matching contacts per company. Set the daily new contact suggestions limit.
3. Selling context. The value proposition for this segment, in the words your best rep uses on calls.
4. Outreach. Sequence outreach or adaptive outreach, and the style of the emails.
5. Guardrails. What the agent may not say or claim. Pricing, competitor names and discounts go here.
6. Automation. How much it does without a rep. Start with drafts that a rep approves.
7. Test, then publish. Use Test, pick a contact, click Generate, read the full sequence, then Review and turn on.

Start with one play for one segment. Two plays that overlap will enrol the same companies and a prospect gets two sets of emails.

What does running HubSpot AI agents cost in a real month?

HubSpot bills per completed action, not per seat, at $0.010 a credit in packs of 1,000 for $10. For a mid-sized team, running both agents usually lands in the hundreds of dollars a month before the credit allowance your subscription already includes. The worked example below uses the published rates in the table above.

A support team handles 1,200 text conversations a month. Assume the Customer Agent resolves 40% of them; that rate is our planning assumption for a first month, not a HubSpot figure. That is 480 resolutions at 50 credits, or 24,000 credits. The sales team sets the Prospecting Agent's daily limit to 20 leads and it runs 22 working days: 440 outreach recommendations at 100 credits, or 44,000 credits. Together, 68,000 credits, about $680 a month.

Your subscription includes a monthly credit allowance that varies by tier; we have not stated a figure here because it differs by account, so read yours in the billing settings. One boundary is worth knowing: HubSpot pauses usage-based features when included credits run out and you have not bought a pack, which makes the allowance the safest place to run the first test. The AI credits page has every rate and the arithmetic for other volumes.

How long does it take, and who does the work?

The agent itself switches on in days. The portal work before it takes two to four weeks for most portals, almost all of it data cleanup. Four people are involved: a HubSpot admin, a support lead who owns the knowledge and handoff rules, a sales lead who owns the plays, and a named reviewer for week one.

The timeline we plan to:

• Weeks 1 to 3: readiness. Deduplicate, write the property dictionary, name the source of truth for each field an agent reads, scope permissions.
• Week 4: activation. Five business days on INSIDEA's Fast Track, ten on Standard: forecast, first cohort, switch-on.
• 24 hours and day 7: reconciliation. Predicted credits against billed credits, escalations read, then a decision on how far to widen it.

If your portal is already clean and the use case is narrow, you can skip to activation and do it yourself from HubSpot's documentation. Plenty of teams should.

When is a HubSpot Breeze agent not enough?

When the job needs data HubSpot does not hold or a decision the packaged agents are not built for. Lead scoring from product usage, support triage across two ticketing systems and call summaries written back to custom objects are custom agent work: built on your infrastructure, triggered by HubSpot workflows, writing back to the record.

The AI agents on HubSpot page sets out both routes side by side, and the Customer Agent playbook goes deeper on the support build.

How INSIDEA adds AI agents to HubSpot

INSIDEA is an Elite HubSpot Partner rated 4.99 across 450+ verified reviews, with an AI and Automation practice that activates HubSpot's own agents and builds custom ones. For Breeze agents we run the readiness check, forecast credits before switch-on, activate on a bounded first cohort and reconcile at 24 hours and day 7: $300 on the Fast Track in five business days or $250 Standard in ten. Custom agents run on a seven-week plan, fixed fee from $2,000. The agent activation page has the rates for every agent.