Agent Studio is GoHighLevel's visual builder for custom AI agents. Since February 2026, workflows can run those agents as a step, which turns them from a chatbot into part of your automation. This guide explains what Agent Studio is, how its parts fit together, how to call an agent from a workflow, and when to use it instead of Conversation AI or an external agent.
1. What Agent Studio is
Agent Studio (under AI Agents → Agent Studio) lets you build "flow agents" on a canvas: you connect nodes that decide how the agent starts, what it looks up, which tools it calls and what it returns. It sits alongside the other AI features in GoHighLevel:
- Conversation AI replies to customers in SMS, chat and social DMs.
- Voice AI handles phone calls.
- Agent Studio builds custom, multi-step agents that can search, call APIs and return results to other parts of GoHighLevel.
It needs the AI Employee add-on with Agent Studio access, and usage is metered like other GoHighLevel AI. For the bigger picture, see the complete GoHighLevel AI automation guide.
2. The building blocks
- Start trigger: what launches the agent. Options include Form Submitted, Lead Tag (added or removed), Chat Message and Workflows.
- AI Agent node: the "brain" that reads the input, understands context and writes the response.
- Search Knowledge Base: answers from your own documents, FAQs and policies.
- Search Web: looks up current public information.
- API Call: sends data to, or fetches data from, another system.
- MCP Server: connects the agent to external tool servers.
- Router: sends the flow down different paths based on conditions or detected intent.
- Sequential: runs a group of steps in a fixed order.
- Data collection nodes: email, phone number, single choice and text input.
- Variables and a Global Prompt: reusable values, plus tone and behaviour rules that apply across the agent.
3. Versions: Staging and Production
Saving and publishing updates the Staging version, so you can change and test an agent without affecting live contacts. When it's ready, you promote Staging to Production, which is the version that runs for real contacts. Use the canvas Test option to simulate triggers before you promote anything.
4. Running an agent from a workflow
The Invoke Agent Studio Agent workflow action is what makes Agent Studio useful for automation. When a workflow reaches this step, the agent runs in the background, returns its response, and the workflow continues with that response available to later steps.
Requirements:
- The agent is promoted to Production (Staging versions can't be selected).
- The agent uses the Workflows trigger.
- The agent responds within 60 seconds, or the step fails.
The action has four fields:
- Agent: the Production agent to run.
- Message: optional instructions, with merge fields for context.
- Input Variables: map workflow values (contact fields, form answers) to the agent's inputs. A missing required input makes the step fail.
- Store Output As: saves the agent's response so later steps can use it in messages, If/Else branches, custom fields, notes or webhooks.
If the step fails, the error appears in the workflow's execution log, which is the first place to look (see how to debug a GoHighLevel workflow).
5. Example: AI lead triage from a web form
- Trigger: a workflow starts when the enquiry form is submitted.
- Invoke Agent Studio Agent: pass the name, service requested and message as input variables. The agent checks the knowledge base for the services you offer and classifies the lead.
- Store Output As: save the response, for example as
lead_triage. - If/Else: branch on the result: urgent leads go to a person by SMS straight away, good-fit leads get a booking link, and poor-fit leads get a polite reply.
- Update fields: write a short summary to a custom field or note so staff see it before they call.
Tip: branches are only reliable if the output is predictable. Tell the agent to answer with exactly one label (for example URGENT, GOOD_FIT or NOT_FIT) and put the summary in a separate step or field. A free-text answer is hard to branch on.
The same pattern works for conversation summaries before a hand-off, enriching new leads, drafting follow-ups for a person to approve, and sending structured results to another system through a webhook.
6. Agent Studio vs Conversation AI vs an external agent
| Use | When |
|---|---|
| Conversation AI | Answering customers, qualifying and booking in SMS, chat and DMs, using GoHighLevel's own features. Setup guide: Conversation AI. |
| Agent Studio | Multi-step logic, web search, API or MCP tools, and AI steps inside workflows, while keeping everything in GoHighLevel. |
| External agent (n8n, LangChain, LangGraph) | Large document sets with RAG, full control of the model and data handling, long-running jobs over 60 seconds, or one agent shared across several systems. It connects to GoHighLevel through webhooks and the API. |
7. Guardrails
- Test before Production: run real examples through Staging, including odd and incomplete ones.
- Ground it: answer from the knowledge base, not the model's general knowledge.
- Plan for failure: decide what the workflow does if the agent step fails or times out, so no lead is dropped.
- Keep a person in the loop: for anything sensitive, have the agent draft and a person approve.
- Messaging rules still apply: AI-written texts need A2P 10DLC in the US and consent under GDPR/PECR in the UK and EU.
Agent Studio is still new and changes often, so check names and options in your account before relying on any guide, including this one.
Need help building it?
I'm Arslan Mumtaz, a software engineer who builds GoHighLevel automation and AI agents, inside GoHighLevel with Agent Studio and Conversation AI, and outside it with n8n, LangChain and LangGraph, for businesses and agencies in the US, UK, Europe and Australia. See AI & automation and pricing, or start with a $97 audit.