How to use the AI Agent action in GoHighLevel workflows
Heads up: the trial and plan buttons on this page use my affiliate links. If you sign up through one I get a commission. It doesn't cost you more.
To use the AI Agent action in GoHighLevel, open a workflow under Automation > Workflows, click + to add an action, pick AI Agent from the Workflow AI section, write plain-English instructions, choose a model and add up to 10 tools it's allowed to use (Send SMS, Add Tag, Search Knowledge Base and so on). On each run the agent decides which tools to call and in what order, and you pay for the tokens it uses plus any premium tools or web searches, which no AI Employee plan covers (as of Nov 10, 2026).
It's one step inside a normal workflow, so the trigger, the waits and the rest of the automation stay exactly as you built them. Below is a full build for a local business, every setting that's been added since it launched in April (knowledge base search, MCP, Claude and Gemini models, skills, web search), what a run really costs, and a straight answer on when to use this vs a Managed Agent or a Conversation AI bot.
30 days free instead of HighLevel's usual 14, plus a 1:1 setup call with my team, weekly live calls and my agency resources from day one. See everything you get →
Short version
- 01Where: Automation > Workflows > + > AI Agent (Workflow AI section). It works with any trigger.
- 02Build: instructions, a model, up to 10 tools, optional skills and web search, then an output format (None, Text or JSON) so later steps can use the answer.
- 03Models: Anthropic, Google and OpenAI, with Low, Medium or High effort on reasoning models. Use a fast model for sorting and tagging, a bigger one for multi-step work.
- 04Cost: tokens at the model's rate, plus premium tools at their normal per-execution price, plus web search at 1 cent a search (OpenAI, Claude) or 1.4 cents (Gemini). AI Employee plans don't include it (HighLevel's AI pricing doc, Nov 10, 2026).
- 05Use it for one smart step inside a workflow you already have. Use a Managed Agent when the whole job is the agent, and Conversation AI when a bot has to chat with leads back and forth.
Build a lead follow-up agent with the AI Agent action, step by step
- 01
Pick one job and the trigger
We're building a new-lead agent for a dental clinic. Trigger: Form Submitted, filtered to the website's appointment request form. The agent reads the answers, checks the service against the clinic's knowledge base, sends one SMS and one email, tags the lead, and hands a clean result back to the workflow. Texting needs approved A2P first (A2P 10DLC registration).
- 02
Build the knowledge base first
AI Agents > Knowledge Base, create one called Clinic Info. Crawl the services page and add an FAQ with hours, insurance they take and what counts as an emergency. The agent pulls only the chunks it needs at runtime, which keeps the prompt short and the token bill lower.
- 03
Add the AI Agent action
Automation > Workflows, open the workflow, click + under the trigger and choose AI Agent from Workflow AI. Rename it ("New lead agent") so the workflow reads well later. You can start from a template like Form Lead Follow Up, which fills in the instructions and tools for you, or pick Build Your Own.
- 04
Write the instructions
HighLevel's doc says good instructions have a role, what trigger data the agent gets, the steps, if/then rules, how to write messages and what never to do. Something like: "You follow up new appointment requests for Bright Smile Dental. Look up the service they asked about in the Clinic Info knowledge base. Send one short SMS and one email that thank them, confirm the service and share the booking link. If they mention pain, swelling or a broken tooth, add the tag lead-emergency and say we'll call within 15 minutes. Never quote a price and never give medical advice." Rough notes are fine too: the button under the field rewrites them into a structured prompt, and you can undo it.
- 05
Pick the model and effort
Models are grouped by provider (Anthropic, Google, OpenAI) and reasoning models show a thinking chip. This job is short and simple, so a fast model on Low effort is plenty. HighLevel's own guidance: GPT-6 Luna for fast, high-volume work like classifying and tagging, GPT-6 Sol for complex multi-step work.
- 06
Add the tools, and lock what shouldn't move
+ Add Tool: Search Knowledge Base (pick Clinic Info, Dynamic query), Send SMS, Send Email, Add Tag. Each tool has a Let AI decide all field values toggle. Turn it off where you want control, for example fix the email's from-name and the booking link, and let the agent write only the message body. Up to 10 tools per agent, and Date Calculator and Math Operations don't count.
- 07
Attach skills
Open Skills below Tools and attach the built-in Sms-writer and Humanizer so texts come out short and don't sound like AI. If the clinic has rules that only apply sometimes, like how to handle an emergency, put them in a custom skill with a clear "When should the agent use this?" line and keep the main instructions short.
- 08
Set the output
Advanced Options > Output Format > JSON, with properties like service (string), emergency (boolean) and summary (string). Leave Conversation Memory off for a one-time follow-up. Turn it on for agents that touch the same contact again and again, like a weekly pipeline check-in.
- 09
Use the answer in the next steps
Add an If/Else on the agent's emergency value. Yes: internal notification to the front desk and a call task due in 15 minutes. No: drop them into the normal nurture. Save the summary into a contact field so whoever calls sees it.
- 10
Test, then read the Execution Logs
Publish, submit the form with your own number and email (it sends real messages), then open the Execution Logs tab and click the AI Agent run. You'll see the input, every model call, every tool call with success or failure, and the tokens used. Try a normal request, an emergency and a half-empty form before a real lead ever hits it.
What the AI Agent action actually does
Normal workflows are fixed: every branch and every field has to be built by hand. The AI Agent action is one step where you describe the goal, hand it tools, and it plans the order and fills in the inputs itself. HighLevel's launch example is a good one: instead of separate steps to find a contact, generate a booking link, format a date and send follow-ups, one AI Agent step does the lot.
It already knows a lot without setup. It can pull contact details, pipeline and calendar data from the CRM, read the contact's past conversations across SMS, email, WhatsApp, Instagram, Messenger and TikTok when it decides they're relevant, and read call transcripts if your instructions mention them. Write "assign the deal to Sarah" and it works out which user Sarah is.
The way I explain any AI agent, "every AI agent has three components." A knowledge base, which is "the brain." The prompt, the "instructions on how the AI agent should behave." And actions, "giving it the ability to do more than speak." In this action those are the Search Knowledge Base tool, the Instructions field and the Tools list. And no, you don't need to code. I studied finance, and like I say in my videos, "if you know how to speak English and you know how to use a computer, you're going to be fine."
AI Agent action vs Managed Agent vs Conversation AI bot: which to use
All three are "AI agents" in HighLevel's menus, and people mix them up constantly. The short version: Conversation AI talks to leads, a Managed Agent is a whole AI teammate with its own triggers, and the AI Agent action is one smart step inside a workflow you control.
My rule of thumb. If a lead texts, DMs or chats and needs a back-and-forth that ends in a booking, that's Conversation AI (full build in Train Conversation AI). If the job is bigger than one step, runs on its own schedule or needs things like browsing and files, build a Managed Agent in Agent Studio (Agent Studio) and call it from a workflow with the Invoke Managed Agent action if you need to. If you already have a workflow and one part of it needs judgment, like reading a form, sorting a lead or writing a personal message, drop in the AI Agent action.
The AI Agent action is also the only one of the three that no AI Employee plan covers, so on a client already paying for Growth or Unlimited, check the volume before you pick it.
| AI Agent action | Managed Agent | Conversation AI bot | |
|---|---|---|---|
| Where you build it | Inside a workflow, as one step | AI Agents > Agent Studio > Managed Agents | AI Agents > Conversation AI |
| What starts it | Any workflow trigger | Its own triggers (forms, tags, opportunities, schedule, chat) or a workflow | An inbound message on a connected channel |
| Best for | One judgment step: sort, enrich, write, decide | A whole back-office job run by the agent | Chatting with leads and booking them |
| Talks to customers? | Through its SMS and email tools, or a shareable chat page | Mostly works in the background | Yes, that's its job |
| Billing | Tokens + premium tools + web searches, pay per use | Per run, pay per use without a plan | Per response, by tokens without a plan |
| Covered by AI Employee Growth ($50) | No | 100 runs a month | 1,000 responses a month |
| Covered by AI Employee Unlimited ($97) | No | 1,000 runs a month | Unlimited, fair use |
From HighLevel's help docs and AI Product Pricing doc (modified Nov 10, 2026), read Nov 10, 2026. Plan prices are per sub-account per month.
What a run costs (and how to keep it low)
HighLevel's formula for every execution: tokens used across all the agent's model calls times the selected model's rate, plus each premium tool it calls at that tool's normal per-execution price. Standard tools like Send SMS, Update Contact Field and Add Tag add no premium charge (the SMS itself is still billed like any text). Premium app tools like ClickUp, Airtable or Notion cost what they cost as standalone actions. Web search is on top: 1 cent a search on OpenAI and Claude models, 1.4 cents on Gemini, only when the agent actually searches.
HighLevel's AI pricing doc (Nov 10, 2026) doesn't publish a separate per-model rate card for this action, so read the real number in the Execution Logs (tokens) and the account's billing page. To get a feel for it, here's HighLevel's own example rate of $1.25 per million input tokens and $10 per million output: a run that reads 20,000 tokens and writes 1,500 costs about 4 cents. Add two web searches on an OpenAI model and it's about 6 cents. That's an illustration, not a quote, so test with your model and your prompt.
Things that keep it cheap: a lighter model and Low effort for simple jobs, the knowledge base tool instead of pasting the whole FAQ into the instructions, only the tools the job needs, and web search off unless the job really needs live facts. HighLevel has also cut tokens on its side twice (Apr 17 and Aug 20, 2026), up to 50% on the same task. The full list of per-use fees is in Usage fees.
| Charge | Price |
|---|---|
| Model tokens | Selected model's per-token rate, every run |
| Standard tools (Send SMS, Add Tag, Update Contact Field) | No premium charge (normal SMS and email fees still apply) |
| Premium app tools (ClickUp, Airtable, Notion) | Same per-execution rate as the standalone action |
| Web search, OpenAI or Claude model | $0.01 per search |
| Web search, Gemini model | $0.014 per search |
| Shareable chat page | Each visitor reply billed like an AI Agent execution |
| AI Employee plans (Starter, Growth, Unlimited) | Not included, always pay per use |
From HighLevel's AI Agent, Web Search and AI Product Pricing docs, read Nov 10, 2026.
Which model to pick
Since Aug 20, 2026 the model dropdown has Anthropic (Claude) and Google (Gemini) models next to OpenAI. So yes, if you searched "Claude in GoHighLevel workflows", this is where it lives. Each model has a one-line description, and the ones that reason show a thinking chip with a Low, Medium or High effort setting.
How I'd choose: sorting, tagging, short replies and summaries get a fast, cheap model on Low effort. Research, multi-step jobs with lots of tool calls or anything where a wrong call costs the client money get a stronger model, and only then raise the effort. Higher effort means more tokens, so don't max it by default. The list changes often, so read the descriptions in the dropdown when you build.
Knowledge base and custom values
The Search Knowledge Base tool lets the agent look things up in one or more of the sub-account's knowledge bases: services, prices, booking rules, policies. Choose a Static query (the same search every time) or a Dynamic query where the agent writes the search itself, and set how many chunks it pulls back. Fewer chunks, fewer tokens.
The Update Custom Value tool lets the agent write to custom values, like the business-hours message for today or a promo code that fits the contact. You can fix which custom value it writes to, or let it pick. It replaces a big If/Else tree that would otherwise set each value by hand.
MCP as a tool: connect outside apps
Since Apr 17, 2026 the agent can use any MCP server as a tool. In the AI Agent action click Add Tools, open the MCP tab and click Add Connection. Fill in a name, the server URL, the transport (HTTP Streamable or SSE) and the auth type (None, Bearer Token, API Key, OAuth2 or Custom Header), then Test Connection. HighLevel lists every tool that server offers and you switch on only the ones this agent should use.
HighLevel's examples are Exa and Tavily for web search and Browserbase for browser automation, and custom servers work too. Two cautions from me: only connect servers you trust with the client's data, and keep the tool list tight, because every extra tool is something the agent can call when you didn't expect it. If you want the other direction, Claude working inside HighLevel, that's Connect Claude to GoHighLevel.
Skills and web search (added Oct 29, 2026)
Skills sit below Tools. A skill is a reusable set of instructions with a name and a "When should the agent use this?" line. On every run the agent only reads those two lines, and loads the full instructions when the run matches, so situational rules don't bloat every prompt. Add one with Create with AI, by writing it yourself, or from your library. Every account gets four built-in skills: Humanizer, Write-well, Cold-email and Sms-writer.
Skills are shared across the account. Edit one and every AI Agent action using it changes, which is great for fixing a rule once and dangerous if you forget. Detach removes it from one action, Delete removes it everywhere. A skill only tells the agent how to do something, so it still needs the tools to act.
Web search lives under Built-in tools, one toggle. The agent searches only when it needs facts it doesn't have, and you can paste URLs into the instructions to point it at pages you trust first. Hover the crown icon in the builder to see the cost. It's for changing public info like rebates or opening hours of a venue. For the client's own facts, use the knowledge base.
How to sell this to a local business
Find the gap first. I tell people to go to the business's site, fill out the contact form and "see how long it takes for someone to actually get back to you. Do they do it instantly? Do they take 6 hours, 24 hours?" If leads sit there, the agent above fixes it: every request gets a personal text and email in under a minute, and emergencies get flagged for a call.
Then sell the outcome: faster replies and no lead left sitting, because the money is made in the follow-ups. Build it once in a clean sub-account, save it in a snapshot with its knowledge base and custom values (GoHighLevel snapshots), and you could build this for a local business and charge a monthly fee you set. Because no AI Employee plan covers the action, price the tokens in or rebill the usage. More on the workflow side in GoHighLevel workflows.
30 days free instead of HighLevel's usual 14, plus a 1:1 setup call with my team, weekly live calls and my agency resources from day one. See everything you get →
Other ways to start
My bonuses come with every one of these.
Best if you're new and want live help setting up
30-day trial + a free live setup bootcamp
Same 30-day trial, plus a seat in HighLevel's free live bootcamp: a click-by-click Zoom class with a HighLevel coach and small breakout rooms, every weekday at 11am CT, 60 minutes max. You set up your account during the class instead of figuring it out alone. HighLevel says you can launch in 48 minutes.
Best if you're not sure which plan you need
Compare all 3 plans first
HighLevel's own pricing page with Starter ($97), Unlimited ($297) and Founder ($497) side by side, monthly and yearly. Every plan there has a 14-day free trial.
Best if you'll resell HighLevel as your own software
Go straight to the $497 Founder plan
HighLevel's top plan (it used to be called Agency Pro), with a 14-day free trial. Everything in Unlimited plus SaaS mode, rebilling with your own markup, Founders Club and HighLevel Partner Network access. For when you're ready to resell the software, not for day one.
FAQ
Where is the AI Agent action in GoHighLevel?
In Automation > Workflows. Open a workflow, click + to add an action and pick AI Agent from the Workflow AI section, or search for "Agent". It works with any workflow trigger.
How much does the AI Agent action cost?
Each run is billed by the tokens the model uses at that model's rate, plus premium tools at their normal per-execution price, plus web search at 1 cent a search on OpenAI and Claude or 1.4 cents on Gemini. Standard tools like Send SMS and Add Tag add no premium charge (as of Nov 10, 2026).
Is the AI Agent action included in the AI Employee plan?
No. HighLevel's AI pricing doc (Nov 10, 2026) lists Workflow AI actions as not included in Starter, Growth or Unlimited. They're always pay per use. Conversation AI and Managed Agents do get allowances on Growth and Unlimited.
Can I use Claude or Gemini in GoHighLevel workflows?
Yes. Since Aug 20, 2026 the AI Agent action's model dropdown has Anthropic and Google models next to OpenAI, with Low, Medium or High effort on reasoning models.
How many tools can one AI Agent action use?
Up to 10. Internal helpers like Date Calculator and Math Operations don't count toward the limit, and web search is a built-in toggle outside the 10.
What's the difference between the AI Agent action and Invoke Managed Agent?
The AI Agent action is built inside the workflow step itself. Invoke Managed Agent hands a task to an agent you built and published in Agent Studio and brings its answer back into the workflow.
How do I see what the agent did?
Open the workflow's Execution Logs tab and click the AI Agent run. It shows the input, every model call, every tool call with success or failure, and the tokens used, in a table or raw JSON.
Sources / last checked Nov 10, 2026
- HighLevel help docs: AI Agent action. Workflow Action - AI Agent (modified Oct 1, 2026) and Web Search in the AI Agent Action (modified Oct 30, 2026), read Nov 10, 2026.
- HighLevel help docs: pricing. AI Product Pricing (modified Nov 10, 2026): AI Employee Starter $20, Growth $50, Unlimited $97 per sub-account per month, Workflow AI actions not included in any plan, Conversation AI and Managed Agent allowances, read Nov 10, 2026.
- HighLevel help docs: other AI tools. Managed Agents, Conversation AI & Ask AI: What's the Difference? (modified Oct 1, 2026), Workflow Action - Invoke Managed Agent (Sep 24, 2026), Invoke Agent Studio Agent in Workflows (May 12, 2026) and AI Tools in HighLevel (Oct 27, 2026), read Nov 10, 2026.
- HighLevel changelog. AI Agent Action in Workflows (Apr 8, 2026), Token Usage Optimization and MCP as a Tool (Apr 17, 2026), Anthropic & Google models (Aug 20, 2026), Shareable Agents in Workflows (Oct 27, 2026), Skills and Web Search in the AI Agent Action (Oct 29, 2026), ideas.gohighlevel.com/changelog, read Nov 10, 2026. No newer entries on this action through Nov 10, 2026.
- My video. My dental speed-to-lead build (Jun 2026) for the three parts of an agent and the after-hours test. Its title and opening have an income figure, so it isn't embedded and the figure isn't repeated.

Author
Hamza Baig / Hamza Automates
Founder, Hexona Systems
I run Hexona Systems on GoHighLevel. HighLevel gave it the SaaSpreneur Platinum award in 2024 and Diamond in 2025 and 2026, and I spoke at their LevelUp event in 2026. Hexona powers 1,000+ agencies, we've served 1,500+ businesses, and I've taught 45,000+ students. Everything on this hub is something I use myself or have shown on camera.
- Platinum SaaSpreneur
- 2024
- Diamond SaaSpreneur
- 2025, 2026
- LevelUp speaker
- 2026
- YouTube subscribers, Oct 2026
- 34,200