GoHighLevel Agent Studio: Build Custom AI Agents (2026)
GoHighLevel Agent Studio: Build Custom AI Agents (2026)
By Dr Priya Jaganathan, GoHighLevel Certified Admin · HL Growth Partner, Australia · Updated 7 August 2026 · 8 min read
Key takeaway: GoHighLevel Agent Studio is HighLevel's builder for custom AI agents — chat, voice and workflow-embedded — that you train on your own knowledge base, point at a specific goal, and deploy across SMS, web chat, WhatsApp and phone. If you've outgrown the basic Conversation AI toggle and want agents that book appointments, qualify leads and hand off to humans on your rules, it's worth adopting. Budget for usage-based billing and set guardrails before you go live.
On this page: What is Agent Studio? · Agent Studio vs Conversation AI vs AI Employee · Building your first agent · Channels and deployment · Workflow actions, guardrails and human handoff · Testing before launch · Pricing, usage billing and rebilling · Common mistakes · FAQ · Related articles
GoHighLevel Agent Studio changed how I build AI for clients. Before it arrived, setting up AI in HighLevel meant flicking on Conversation AI, pasting a personality prompt into a single box, and hoping the bot stayed on script. Agent Studio replaces that with a proper build environment: named agents, defined goals, trained knowledge bases, channel assignments and workflow-level control over what the agent can actually do in your CRM.
I run a HighLevel agency from Australia and manage sub-accounts for trades, allied health and professional services clients. In this guide I'll walk through what Agent Studio is, how it differs from Conversation AI, Voice AI and the AI Employee bundle, how to build and test an agent properly, and how the usage billing works when you're rebilling clients through SaaS Mode. No hype — just what I'd tell you across the table.
What is GoHighLevel Agent Studio?
Agent Studio is the section of HighLevel (under the AI Agents area of your sub-account) where you create, train and manage individual AI agents. Each agent is a self-contained unit with its own name, personality prompt, goal, knowledge base, channel assignments and action permissions. You can run several agents in one sub-account — a reception agent answering web chat, a reactivation agent working old leads over SMS, and a voice agent answering the phone after hours — each trained differently and each measurable on its own.
That's the core shift: from "one bot per account" to "a team of purpose-built agents". A HighLevel Agent Studio build treats each agent like a staff member with a job description, not a chat widget with a personality.
The three agent types
In practice you'll build three kinds of agents. Conversation agents handle text — SMS, web chat, WhatsApp, Facebook and Instagram DMs, all through the LeadConnector infrastructure. Voice agents answer or place phone calls, transcribe them, and log outcomes against the contact. Workflow agents sit inside your Workflows as actions — they don't chat with anyone; they make decisions, summarise, categorise or draft content mid-automation, which is where some of the most valuable and least flashy wins live.
Agent Studio vs Conversation AI vs AI Employee
This is the question I get most from agency owners, because HighLevel's AI naming has sprawled. Here's how the pieces relate.
| Feature | Conversation AI (classic) | Agent Studio | AI Employee |
|---|---|---|---|
| What it is | Single AI bot setting per sub-account | Builder for multiple named, trainable agents | Bundle of all HighLevel AI tools at one flat price |
| Number of agents | Effectively one | Multiple per sub-account | Multiple (includes Agent Studio) |
| Channels | SMS, chat widget, social DMs | SMS, web chat, WhatsApp, social, voice calls | All of the above |
| Training | Prompt plus limited Q&A | Goals, prompts, knowledge base per agent | Same as Agent Studio |
| Workflow integration | Basic on/off in bot settings | Agents as workflow actions and triggers | Same as Agent Studio |
| Billing model | Per message usage | Usage-based (messages and voice minutes) | Flat monthly fee, unlimited usage |
| Best for | Simple FAQ and booking bot | Custom multi-agent builds | Heavy users who want cost certainty |
The short version: Agent Studio is the tooling; AI Employee is the pricing wrapper. If an account uses AI lightly, pay-as-you-go usage through Agent Studio is cheaper. Once usage climbs, the flat-rate AI Employee plan usually wins — I've broken down the maths in my guide to GoHighLevel AI Employee features and ROI. And if you've already got a classic bot running, my GoHighLevel Conversation AI guide covers migrating that setup across without losing your training data.
Building your first agent: goals, prompts and knowledge base
Start with the goal, not the prompt
Every agent in Agent Studio has a primary goal — book an appointment, qualify a lead, collect specific details, or answer support questions. Pick one. Agents with a single goal convert dramatically better than agents asked to "handle everything". For a Brisbane plumbing client, our web chat agent has exactly one goal: capture the job type, suburb and urgency, then book a slot on the emergency or standard calendar. Nothing else. It books over 60% of after-hours enquiries without a human touching them.
Write the prompt like an SOP
The personality prompt is where most builds fall over. Treat it like a standard operating procedure for a new hire: who the business is, tone of voice (Australian spelling and idiom matters — nobody in Parramatta wants to be told about their "estimate for the fall"), what the agent must always do, what it must never do, and exactly when to stop and hand off. Short, declarative rules beat long flowing paragraphs.
Train the knowledge base properly
Agent Studio agents pull answers from a knowledge base you control — uploaded documents, FAQ entries and crawled website URLs. Feed it your real pricing structure, service areas, cancellation policy and warranty terms. Then prune it: outdated PDFs with old prices are the number one cause of agents confidently quoting wrong figures. Review the knowledge base whenever the client changes pricing, and re-crawl the site after major updates.
Give the agent structured data to work with
Map the details your agent collects to custom fields on the contact, and have it apply tags as the conversation progresses — ai-qualified, budget-confirmed, wants-callback. Those tags and fields become the fuel for your pipelines and follow-up Workflows, and they're what makes the agent's work visible to the humans in the account.
Channels: SMS, web chat, WhatsApp and voice
Once an agent is trained, you assign it channels. A single agent can cover multiple text channels with the same brain, which keeps behaviour consistent whether the lead lands via the website chat widget or an Instagram DM.
Text channels
SMS remains the workhorse in Australia — open rates are unbeatable, and Agent Studio agents reply on your LeadConnector number within seconds. Web chat converts anonymous traffic into contacts with mobile numbers, so the conversation survives the visitor leaving the page. WhatsApp requires the WhatsApp integration on your sub-account and is increasingly the preferred channel for migrant-community-heavy service businesses I work with in western Sydney and Melbourne.
Voice agents
Voice agents answer inbound calls on a dedicated or after-hours number, speak with configurable Australian-accented voices, and log a transcript plus summary to the contact record. Set them up honestly — the agent should identify itself as an AI assistant — and configure call transfer rules so genuine emergencies reach a human. Voice is billed per minute, so it's also where usage costs bite hardest; my GoHighLevel Voice AI guide goes deep on call flows, transfer logic and minute costs.
Workflow actions, guardrails and human handoff
Agents inside Workflows
This is where Agent Studio stops being a chatbot feature and becomes infrastructure. Agents plug into Workflows in both directions. Events like "appointment booked by AI" or "agent goal completed" can fire GoHighLevel workflow triggers, kicking off confirmations, pipeline stage moves and internal notifications. And workflow agent actions let an agent make a decision mid-automation: summarise the conversation into a note, score the lead's intent, choose a branch, or draft a personalised reply for a human to approve. One allied health client uses a workflow agent purely to read inbound form submissions and route them to the right practitioner's pipeline — it never speaks to a patient, and it saves the admin team about an hour a day.
Guardrails
Guardrails are the boundaries you set on agent behaviour, and they're not optional in my builds. The essentials: restrict answers to the knowledge base so the agent says "I'll check with the team" instead of inventing an answer; block topics it must never discuss (medical advice, legal outcomes, competitor comparisons); cap message counts per conversation so a confused agent can't loop; and constrain actions — an agent that can book calendars doesn't also need permission to edit opportunities.
Human handoff
Define explicit handoff conditions: the contact asks for a human, mentions a complaint or refund, or the agent fails to reach its goal within a set number of exchanges. On handoff, pause the agent on that conversation, notify the right user inside HighLevel (or via SMS), and add a tag like needs-human so nothing slips. The businesses that win with AI aren't the ones that remove humans — they're the ones where the AI does the triage and the humans handle the moments that matter.
Testing before you go live
Agent Studio includes a built-in test chat — use it, but don't stop there. My pre-launch checklist: run at least 20 realistic conversations, including hostile ones ("your competitor is cheaper", "I want a refund", profanity, off-topic rambling). Test on the real channels, because SMS truncation and WhatsApp formatting behave differently from the test panel. Verify every booking lands on the correct calendar with the correct duration. Check tags and custom fields are actually being written. Then soft-launch: run the agent live on after-hours enquiries only for a week, review every transcript, tighten the prompt, and only then let it handle business-hours volume. HighLevel's own documentation at help.gohighlevel.com is genuinely good on the mechanics if you get stuck on a setting.
Pricing, usage billing and rebilling for agencies
How the billing works
Agent Studio itself is included with HighLevel plans (see current tiers at gohighlevel.com/pricing); what you pay for is usage — per message for text agents and per minute for voice agents, drawn from the sub-account's wallet with auto-recharge. Alternatively, the AI Employee unlimited plan (around US$97/month per sub-account) covers all AI usage at a flat rate. My rule of thumb: light usage stays pay-as-you-go; once a sub-account's AI usage bill approaches roughly half the flat rate for two consecutive months, switch it to unlimited.
Rebilling through your agency
For agencies, this is a margin line. With rebilling enabled in your agency settings, you can mark up AI usage costs to clients — and in SaaS Mode you can bake AI agents into your plan tiers, with usage rebilled automatically at your multiplier through the client's wallet. Two practical tips from my own sub-accounts: first, set the client's expectations on voice minutes in writing, because a chatty voice agent on a busy line adds up quickly. Second, build your trained agents, prompts and supporting Workflows into a snapshot per niche — a proven "AI receptionist" snapshot for trades or clinics becomes a productised offer you can deploy to a new sub-account in an afternoon. Pair it with GoHighLevel Content AI for on-brand copy across the same accounts and your AI stack becomes a genuine recurring revenue product, not a novelty.
Common mistakes to avoid
- One agent, every job. Asking a single agent to sell, support, book and reactivate produces a mediocre generalist. Build separate agents with single goals.
- Stale knowledge bases. Old pricing PDFs and outdated service lists make agents confidently wrong. Audit the knowledge base monthly and after every price change.
- No handoff rules. An agent that argues with an angry customer for fifteen messages does brand damage a human would never do. Define escape hatches before launch.
- Skipping channel-level testing. The test panel isn't SMS. Character limits, link previews and WhatsApp formatting all behave differently in the wild.
- Ignoring usage economics. Agencies that never review wallet burn either bleed margin or overcharge clients. Review usage monthly and move heavy accounts to the unlimited plan.
- Launching without transcript reviews. The first two weeks of real conversations are your best training data. Read them and feed the fixes back into the prompt.
If you want a custom AI agent built and trained for your business, book a strategy call with the HL Growth Partner team.
Frequently asked questions
What is GoHighLevel Agent Studio used for?
GoHighLevel Agent Studio is used to build, train and deploy custom AI agents inside a HighLevel sub-account. Agents can chat over SMS, web chat, WhatsApp and social DMs, answer phone calls as voice agents, or run inside Workflows as decision-making actions — each with its own goal, prompt and knowledge base.
Is Agent Studio the same as Conversation AI?
No. Conversation AI is the older single-bot feature with one prompt per sub-account. Agent Studio is a full builder that supports multiple named agents, per-agent knowledge bases, voice capability and deeper workflow integration. Existing Conversation AI setups can be rebuilt as Agent Studio agents, usually with better results.
How much does Agent Studio cost?
The builder is included with HighLevel subscriptions; you pay for usage — per message for text agents and per minute for voice agents, charged to the sub-account wallet. Heavy users can switch to the flat-rate AI Employee plan (around US$97/month per sub-account) for unlimited usage, and agencies can rebill either model to clients with a markup.
Can Agent Studio agents hand conversations to a human?
Yes. You define handoff conditions — such as the contact requesting a person, raising a complaint, or the agent failing its goal after a set number of messages. The agent then pauses on that conversation, notifies your team inside HighLevel, and tags the contact so a human picks it up.
Do Agent Studio agents work with Australian phone numbers and accents?
Yes. Text agents run on your LeadConnector numbers, including Australian SMS numbers, and voice agents offer Australian-accented voice options. Standard Australian telecom compliance still applies, so register your numbers properly and have voice agents identify themselves as AI assistants.
