HighLevel

Voice agents act inside HighLevel's own CRM, creating and updating contacts, booking into one calendar or routing across several, creating opportunities, assigning tasks, sending follow up SMS or email and triggering workflows from the call outcome. Custom Actions send POST webhooks to outside systems mid call, with headers, Bearer or Basic authentication and parameters pulled from what the caller says, such as an order number for a status lookup. Agent Studio adds API and MCP tool nodes, and calls run on LC Phone or Twilio numbers.

Sourcehelp.gohighlevel.com/support/solutions/articles/155000005461-voice-ai-custom-actionsread 2026-10-18

Workflows place outbound Voice AI calls from forms, tags, appointments or pipeline changes, at up to 5,000 calls a day per location and 10 a minute, scheduled between 8 AM and 8 PM in each contact's time zone. One agent can handle both inbound and outbound calls, with actions split between those confirmed during the call and those applied silently afterward. Managed Agents build complete CRM agents from a plain language description, and flow agents in Agent Studio chain LLM, API, MCP, knowledge base and web search nodes with branching and variables.

Sourcehelp.gohighlevel.com/support/solutions/articles/155000006598-voice-ai-outbound-callingread 2026-10-18

Each voice agent can be given its own knowledge base along with a trigger prompt that says when to search it, and edits to the knowledge base reach the agent at once. Knowledge bases take uploaded documents, structured tables and website links that refresh automatically, and they can be bundled with agent templates. Custom values from the account fill agent prompts, and Agent Studio adds a web search node for current information.

Sourcehelp.gohighlevel.com/support/solutions/articles/155000005266-knowledge-base-integration-for-voice-ai-agentsread 2026-10-18

Conversation AI runs Off, in Suggestive mode, where it drafts replies for the team to review and send, or on Auto Pilot, and it can hand over to a human or stop itself. Voice agents transfer calls to people under conditions the business defines and keep to set working hours. Granular permissions decide who can view or change agent goals and settings, Agent Studio changes go to Staging before anyone promotes them to Production, and outbound calling needs location level terms accepted with consent left to the business.

Sourcehelp.gohighlevel.com/support/solutions/articles/155000001022-auto-pilot-mode-in-conversation-ai-for-efficient-communicationread 2026-10-18

Annual SOC 2 Type II assessments cover HighLevel's controls for security, availability and confidentiality, and an UpGuard trust center is linked from its security page. Healthcare clients can sign a Business Associate Agreement under the $297 a month HIPAA package, and data is encrypted in transit with TLS 1.2 or later and at rest with AES 256.

Single sign on, two factor enforcement and granular user permissions decide who can view or change agent goals, conversations and settings, with separate Voice AI and Conversation AI permissions, and audit logs record what users did. Private Integration Tokens carry scopes that limit what an MCP client or API integration can read or change, and customer data is logically separated by tenant.

Sourcehelp.gohighlevel.com/support/solutions/articles/155000000574-highlevel-security-and-compliance-overviewread 2026-10-18

A Voice AI dashboard tracks inbound and outbound call volume, triggered actions, sentiment trends and comparisons over time, and call logs keep transcripts, recordings and outcomes for every live and test call. Performance summary emails go out on a schedule, Conversation AI writes a traceable log entry for each action a bot takes, filterable in the Conversations panel, and AI usage can be reviewed for rebilling. Call logs and transcripts are also available through API and real time webhooks.

Sourcehelp.gohighlevel.com/support/solutions/articles/155000004693-voice-ai-agents-dashboard-overviewread 2026-10-18

Each call and chat leaves its mark on the CRM contact. The agent creates or updates the name, email, address and other fields, and transcripts and summaries stay on the record for the team. Conversation AI sends the conversation history and contact details to the model with every reply, so a lead who first wrote on Facebook and later texts is answered with that earlier exchange in view.

That continuity comes from the business record itself: there is no separate agent memory with its own scope and lifetime that a team could review or clear for one customer without touching the contact, and Managed Agents and flow agents hold runtime and global variables only for the length of a run.

Sourcehelp.gohighlevel.com/support/solutions/articles/155000003911-overview-of-voice-ai-agentsread 2026-10-18

HighLevel runs as a multi tenant service on Google Cloud and AWS with daily backups in the local region, and there is no self hosted edition or customer choice of region. Agencies resell the platform and AI Employee under their own brand, with white label domains, a desktop app, SaaS mode on Agency Pro and rebilling with or without markup, and agents reach callers on LC Phone or Twilio numbers and websites through the chat widget.

Sourcehelp.gohighlevel.com/support/solutions/articles/155000000574-highlevel-security-and-compliance-overviewread 2026-10-18

The Marketplace lists ready to install Conversation AI and Voice AI agent templates that bring their workflows, calendars, custom fields and knowledge bases with them, and agencies can publish and sell their own. Voice AI settings travel in snapshots so a working agent can be copied across client accounts, and Managed Agents can be cloned and shared the same way. Prebuilt roles such as an AI receptionist and AI scheduler cover the common front desk jobs.

Sourcehelp.gohighlevel.com/support/solutions/articles/155000005555-how-to-use-the-ai-agent-marketplace-templates-for-conversation-ai-and-voice-ai-automationread 2026-10-18

Voice agents answer inbound calls around the clock or within set working hours, place outbound calls when a workflow event fires, and also run in a website chat widget. Conversation AI covers SMS, Facebook, Instagram, WhatsApp and live chat, and Managed Agents start from tag changes, form and survey submissions, Facebook lead forms, appointments or chat, or on their own on a one time, recurring or cron schedule. Each Voice AI agent can serve its own phone number or a number pool.

Sourcehelp.gohighlevel.com/support/solutions/articles/155000008245-scheduled-triggers-for-managed-agents-in-highlevelread 2026-10-18

A business can put its voice agent on an external model of its choosing, from Google's Gemini 2.0 and 2.5 Flash family through OpenAI's GPT 4o, 4.1 and 5 lines to Anthropic's Claude Sonnet and Haiku, each billed at its own token price, and pair it with text to speech from OpenAI, Cartesia or ElevenLabs. Speech to speech mode swaps that pipeline for a single realtime model, Gemini Live at $0.10 a minute or OpenAI's realtime model at $0.20, which handles tone and language switches with less delay. Conversation AI and Agent Studio draw on OpenAI's GPT 4.1 and GPT 5 families, priced per token.

Sourcehelp.gohighlevel.com/support/solutions/articles/155000006652-ai-product-pricingread 2026-10-18

An MCP server at services.leadconnectorhq.com/mcp authenticates with a Private Integration Token and location ID and exposes tools over contacts, conversations, messages, calendars, opportunities, payments, locations and custom fields to clients such as Cursor, Windsurf and Claude compatible apps. Voice AI public APIs create, update and delete agents and their actions, list voices and pull call logs and transcripts, a Conversation AI API covers the chat side, and the V2 REST API with OAuth 2.0 and webhooks reaches the rest of the platform.

Sourcehelp.gohighlevel.com/support/solutions/articles/155000005741-how-to-setup-and-use-the-highlevel-mcp-serverread 2026-10-18

Prompt Optimizer opens from the Voice AI agent builder, clones the live agent so production stays untouched, and generates test scenarios from its prompt, language, direction and configured actions.

It then places real voice calls against the clone, can repeat a scenario several times to check consistency, and returns AI evaluation reasoning on whether the agent completed the expected behavior before proposing prompt changes.

Single Web Call and Phone Call tests keep their own logs with transcript and outcome, Conversation AI has an optimizer of its own, Agent Studio offers a test and debug view with Staging ahead of Production, and Managed Agents get 30 free test runs a month in a sandbox before they go live.

Sourcehelp.gohighlevel.com/support/solutions/articles/155000007781-voice-ai-prompt-optimizerread 2026-10-18

Agent Studio's web search tool sends a query to Tavily at a cent a search and hands the results back to the agent as outside knowledge, and knowledge bases pull in website links on a refresh schedule to ground answers. Neither opens a page for the agent to click, type into or move through. Voice agents, Conversation AI and Managed Agents reach other systems through CRM actions, webhooks, API calls and MCP tools, so a supplier portal with no API, or a form on someone else's site, stays out of their reach.

Sourcehelp.gohighlevel.com/support/solutions/articles/155000007965-how-to-use-web-search-tool-in-agent-studio-for-real-time-responsesread 2026-10-18