AI Employee

Have AI on the backside of high level just booking recurring appointments for customers based on parameters set in custom fields. This eliminates manual reschedulings for a in-house scheduler and fixes the issues that everyone has with recurring appointments not having appointment automations retrigger. I was thinking of this more on a couple visits a year scale but it could definitely be made so it would work for whatever number of appointments for whatever time frame. It would even be fine if this had to be triggered by an automation. That's how I'd prefer to have it set-up so that way when the maintenance visit is complete I mark it as completed and then it triggers to the bot to see if anything is on the calendar in the coming months. If not then it goes in and books the appointment. Example: HVAC Maintenance Plan is 1-2 visits a year. Custom Fields for: Which plan, active/inactive, visit 1's month, visit 2's month, day/time preference, and any other parameters you would want to setup. AI sees they are on the "Gold" Plan with 2 visits a year. Today is July 1st and according to visit 1s schedule they should have something for August but nothing is on the schedule. AI sees that, and books them for an open slot on the Maintenance calendar in that month without texting the customer. Appointment booked automations will notify them of the selected day.

It’s crazy that in 2025 the AI still can’t send a simple image or PDF. Businesses need to share promos, price lists, menus, or videos, and text alone isn’t enough. Platforms like ManyChat have allowed this for years. If we want AI to really replace human conversations, it must be able to send media files—just like a real agent. 👉 JPG, PDF, MP4 support is essential for modern sales and service.

Problem Currently, GoHighLevel Managed AI Agents can generate content but lack the ability to properly analyze files sent by customers during conversations. When a customer sends an image, video, audio recording, or document attachment through SMS, WhatsApp, Live Chat, or other supported channels, the AI cannot interpret its contents and respond intelligently. This significantly limits the usefulness of AI agents in real customer interactions. Proposed Solution Introduce multimodal file understanding for Managed AI Agents, allowing them to automatically access, analyze, and understand customer attachments. Requested Capabilities 1. Image Understanding Analyze photos, screenshots, and graphics. Identify objects, products, and visual details. Extract text using OCR. Answer questions about images. 2. Video Understanding Analyze uploaded videos and their contents. Understand scenes, actions, and visual context. Transcribe and interpret spoken dialogue. Summarize videos and answer questions. 3. Document Understanding Read PDFs, Word, Excel, PowerPoint, and text files. Extract information from tables and forms. Summarize, compare, and answer document-related questions. Process scanned documents. 4. Audio Understanding Transcribe voice notes and audio files. Understand spoken requests. Identify relevant information and respond accordingly. 5. Automatic Attachment Processing Automatically detect incoming attachments. Analyze files without requiring manual intervention. Combine file contents with conversation history. Support multiple attachments in one message. Allow agents to use extracted information in workflows and CRM updates. Example Use Cases Hair Salon: Customer sends a hairstyle photo; AI identifies the style and recommends the appropriate service. Real Estate: Customer sends property photos; AI analyzes visible features. Education: Student uploads a transcript; AI extracts relevant academic information. Customer Support: Customer sends a screen recording; AI identifies the reported issue. Expected Outcome Managed AI Agents should not only generate images, videos, and documents but also understand files received from customers . This would enable truly multimodal, context-aware AI conversations and significantly reduce human intervention.

Enable GoHighLevel Managed AI Agents to discover, connect, configure, and create Model Context Protocol (MCP) integrations simply through natural language prompts, similar to ChatGPT. Users should not need to manually configure MCP servers, APIs, or tool definitions. Proposed Solution Allow users to instruct Managed Agents to connect themselves to external applications or build custom MCP integrations. Example prompts: "Connect yourself to Google Drive using MCP." "Find and connect an MCP server for GitHub." "Connect to this MCP server: https://example.com/mcp ." "Read this API documentation and build a custom MCP integration." "Connect yourself to our inventory system using its API." Key Features 1. Prompt-Based MCP Connections Discover and connect compatible MCP servers. Automatically configure tools and permissions. Support OAuth and API key authentication. Test connections before activation. 2. Custom MCP Creation Generate MCP integrations from API documentation. Automatically create tool schemas and configurations. Deploy through supported hosting. Modify and troubleshoot integrations through prompts. 3. Agent-Initiated Discovery Agents should recognize missing capabilities and suggest integrations. Example: User: "Create a task in ClickUp." Agent: "I need ClickUp access. Would you like me to connect its MCP integration?" Once authorized, the agent connects and completes the task. 4. Centralized MCP Management Share connections across Managed Agents. Manage integrations from a central MCP hub. Control agent-level permissions. Enable or disable individual tools. 5. Security Secure credential storage. User approval for connections. Granular permissions and audit logs. Why This Matters Managed Agents should be able to expand their capabilities without requiring developers or complex manual configurations. The goal: If users can describe the capability they need, their Managed Agent should be able to discover, connect, or build the necessary MCP integration through conversation.

The AI would stay on the line to announce the caller and give the called party the option to take the call or reject it. Ideally a rejected caller could leave a message instead.

It would be good to be able to combine the Voice AI and Conversation (text) together on one app that can be placed as a widget on a website just like the Chat Bot widget is now. Give an option to either talk to an agent or text with an agent.

I'd like to request the ability for AI token usage within HighLevel to authenticate and connect using our existing $20/month personal subscriptions from major AI providers like ChatGPT, Gemini, and Claude — instead of requiring separate API key purchases or GoHighLevel's built-in token credits. This would work similarly to how OpenClaw and VS Code's AI integrations operate: you log in with your existing account (OAuth or session-based authentication), and the platform taps into the token limits you already have from your subscription. This means if I already pay $20/mo for ChatGPT Plus, $20/mo for Gemini Advanced, and $20/mo for Claude Pro, I should be able to leverage those token allowances directly inside HighLevel's AI Employee and other AI features. Current pain point: Right now, agencies and power users have to either pay for separate API access through OpenAI/Anthropic/Google (which charges per token on top of subscriptions), or burn through HighLevel's token credits that come with the platform. This creates unnecessary double-dipping costs. The proposed solution would let users: Connect their existing ChatGPT Plus, Gemini Advanced, and Claude Pro accounts via OAuth/login Have AI Employee and other HL AI features pull from those existing token pools Avoid paying twice — once for the subscription and again for API usage This is a huge value-add for agencies running HighLevel at scale. It reduces operational costs, makes the platform more competitive with tools like OpenClaw, and gives users flexibility to use the AI models they're already paying for. It's the kind of integration that would make HL a no-brainer for agencies already invested in these AI subscriptions.

Separate the options for the AI Employee to stop when I send a message from when a workflow is activated. For example, I want it to stop when I intervene in the conversation, but not when a workflow for sending images is activated.

Ability to train a bot for Email channel replies, as Suggestive only to assist team members with ideas of what to say.

There should be a way for the AI Agent to look at zip codes, and addresses to determine how close one appointment is from another. That way it can ensure that it books appointments and routes them in the most effective way that it can. That way it doesn't book one appointment 15 minutes away and the next 1.5 hours away. This would allow for a seamless booking experience and act more like a person actually would and could if they were looking at it. Even if it was just as simple as the appointment title needs to have the zipcode in it so that the AI can read that, that would be fine and could easily be added with current automation schedules