Skip to content
ConsultEvo

HighLevel Release Radar September 7, 2026: Setup and Operations Guide

HighLevel announced five capabilities for the September 7-11, 2026 release week: GPT Live 1 speech-to-speech Voice AI, speech-to-speech outbound calls, mobile call transfers to staff without personal phone numbers, Apple and Google Wallet passes for eligible in-person event tickets, and parallel Ask AI conversations. The practical question is not only what changed, but where each feature belongs in an operating process and what proves that it worked.

This guide translates the announcement into configuration and rollout procedures. It is not a claim that every account had access during release week. The release post is a weekly summary, while current HighLevel help articles provide more detailed setup instructions and constraints. Confirm availability, permissions, location eligibility, and current interface steps in your account.

Use the same operating pattern for each feature: define the job, required input, and owner; configure the feature; then test the success result and the exception route. These capabilities do not share a common integration or data API.

Feature Best-fit job Main control Human fallback
GPT Live 1 Live conversation with configured actions Separate Speech and Reasoning instructions Voice AI or operations owner
S2S outbound Eligible, consented campaign calls Business eligibility gate Campaign or compliance owner
Mobile transfer Hand a live call to staff Eligible staff destination Named escalation owner
Wallet passes Ticket delivery for in-person events Ticket-level attendee and link match Event operations
Ask AI chats Independent research tasks One task per conversation and review User or designated reviewer

Official overview: HighLevel Release Radar for September 7-11, 2026.

Configure GPT Live 1 around two distinct jobs

GPT Live 1 is documented as a full-duplex speech-to-speech Voice AI model. The agent can listen while speaking, and callers can interrupt naturally rather than waiting for a conventional turn-ending signal. HighLevel separates live conversational behavior from more complex reasoning and configured actions. That separation may reduce pauses during supported operations, but it does not guarantee that every tool action will execute concurrently or without delay.

  1. Go to AI Agents > Voice AI > Agent List, then open an agent or create one.
  2. Select GPT Live 1 in the model selector if it is available for the account.
  3. Write Speech model instructions for persona, tone, style, and straightforward spoken conversation.
  4. Write Reasoning model instructions for tools, conditions, and multi-step business rules. Review every configured action before allowing consequential work.
  5. Test with Web Call or Phone Call. Exercise an ordinary question, an interruption, and a request that invokes a configured action. Review the transcript, delegation, and tool activity.

For an appointment request, the Speech model can clarify the caller’s preferred time while the Reasoning model applies booking rules and invokes the configured action. Define required fields, allowed values, identity checks, authorization requirements, and the owner of an uncertain result. Keep the confirmed action record in the system responsible for the booking. A transcript is context, not proof that the booking exists.

Decision point

Treat the spoken response as conversation and the confirmed tool result as evidence. Before updating a CRM record or telling a customer that an action is complete, validate the result returned by the configured action.

A useful operating chain is caller request, live Speech-model conversation, configured Reasoning-model delegation, confirmed action result, validation, and then a destination such as the booking system or a human reviewer. If a required field is missing, a value is outside the allowed set, identity is uncertain, or the action result is ambiguous, stop the write and route the case to the Voice AI administrator or relevant operations owner.

Use HighLevel’s GPT Live 1 setup and testing instructions for current model controls. Teams defining prompts, permitted actions, and human review can also review AI agent design and implementation.

Add a consent gate before speech-to-speech outbound calls

HighLevel documents S2S agent selection in the Voice AI Outbound Call workflow action. The action places a call; it does not decide whether a particular contact may be called. HighLevel states that it does not perform platform-level contact-consent validation before placing these calls. Acceptance of location-level terms is not individual contact consent.

Before the call action, use deterministic business rules for consent evidence, DND and opt-out status, phone format and supported country, campaign suppression, recent-call frequency, location eligibility, and the contact’s local calling time. Fail closed when a required value is missing. AI can interpret free-text intent or summarize a call, but it should not be the sole gate for legally sensitive or irreversible eligibility decisions.

01Trigger and identifyA workflow trigger selects a contact. Confirm that the phone number and contact record are the intended ones, and capture the workflow execution or attempt identifier if the selected reporting source exposes one.
02Apply the eligibility gateCheck recorded consent, DND, opt-outs, suppression, number and country, local time, recent-call rules, and location eligibility. Route blocked contacts to the campaign owner rather than the call action.
03Configure and publishIn the workflow, add Voice AI Outbound Call, choose an eligible agent and From Phone Number, and select an S2S agent where available. Accept location terms if prompted, then publish.
04Test and reviewTest with an appropriate contact, then inspect AI Agents > Voice AI > Dashboard & Logs > Outbound. The campaign owner resolves workflow failures, while consent questions go to the designated compliance owner.

This sequence is editorial operating guidance around documented HighLevel controls, not a supplied HighLevel template. HighLevel documents up to 5,000 calls per eligible location per day and up to 10 per minute per location. Additional safeguards include one call per phone number per day, up to 14 calls per number within 14 days, an 8:00 a.m. to 8:00 p.m. window based on the contact’s phone-number timezone, and supported same-country domestic calling. Check the current outbound calling documentation for eligibility and restrictions.

The same agent can support inbound and outbound calls when configured appropriately for both. Test outbound disclosure, greeting, verbal opt-out, call objective, transfer conditions, and after-hours behavior separately. Do not assume S2S is universally faster or cheaper. Current charges may include model, TTS, LLM, and phone-system components. See HighLevel’s current AI pricing information.

If an external call ledger is part of the design, use one raw row for one outbound call attempt, not one contact or campaign. Proposed fields may include workflow execution ID, action instance ID, contact ID, attempt time, consent decision, and observed call status. These are design suggestions, not verified HighLevel fields. If concurrent workers can process the same item, use a database-enforced unique key and transactional upsert. A lookup followed by create is not safe as the only deduplication control.

Transfer a mobile call to a staff member, not a specific device

In the HighLevel or LeadConnector mobile app, make or answer a call, tap Transfer, open Staff Members, and select an eligible person. Choose Hold and Call for a warm transfer or Transfer and End for a blind transfer. The receiving staff member accepts the call.

A personal phone number is not required, but the staff member must have at least one eligible destination configured. HighLevel lists Web App, Mobile App, Phone, and SIP desk phone as possible destinations. The transfer selects the staff member, not a particular device, so multiple configured destinations may ring. A staff member with no eligible destination appears disabled. Transfer context is shown to the receiver, with display details varying by platform.

Before launch, test both transfer modes on the platforms your team uses. Confirm that the receiving person can identify the caller and transfer context. Since an already transferred call cannot currently be transferred again, name a manual escalation owner for unavailable staff or a caller who needs another handoff. See the mobile transfer setup and limitations.

Deliver wallet passes at the ticket level

HighLevel documents Apple Wallet and Google Wallet passes for eligible in-person event tickets, not online events. Open the event, configure ticket options and registration confirmation notifications, then add {{event.ticket.wallet_links}} to supported email, SMS, or WhatsApp content. Wallet badges may also be available on ticket PDFs.

Each ticket receives its own wallet link and pass. A pass can include the event name, date and time in the event timezone, venue, ticket type, attendee name, and event check-in QR code. For a two-ticket order, validate each ticket separately. The attendee, ticket type, wallet link, and QR code must belong together. An order ID alone is not a suitable ticket-level identity because one order can contain multiple attendees.

Before sending multi-ticket confirmations
  • Confirm each ticket is associated with the expected attendee and ticket type.
  • Render and inspect the wallet link for each ticket in the correct event confirmation context.
  • Scan each ticket’s check-in QR and confirm that it resolves to the expected attendee and ticket.
  • Ensure a retry cannot substitute one attendee’s link for another’s, and name event operations as the owner of mismatches or check-in exceptions.

For an external delivery log, use one raw row for one ticket delivery attempt. A proposed key could combine the event identifier, ticket identifier, channel, and template version, but only if those identifiers are actually exposed by the selected system. Enforce uniqueness in the database or use an atomic upsert when concurrent processing could duplicate a send. This is an implementation recommendation, not a HighLevel-published schema. The wallet pass documentation describes the supported workflow and merge field.

Use Ask AI parallel chats for independent work

Ask AI supports up to 10 parallel conversations, each with separate context and background state. This is a limit on conversations, not a general-purpose API queue. Open Ask AI and submit a prompt; while it runs, open Chat History, start a new chat, and switch between sessions. Work can continue while the panel is closed, the user navigates elsewhere, or the user switches accounts. HighLevel documents completion, failure, and input-needed notifications. An input-needed conversation may pause until the user responds.

Use separate conversations for independent, read-only work such as summarizing two documents or comparing separate reports. Give each conversation one specific task and assign a reviewer for consequential output. Check the result against its source before customer communication, CRM changes, or an operational decision. If two tasks could change the same CRM record, serialize the work or have a person reconcile it. Parallel conversations do not provide transactional conflict control.

For an external review ledger, use one raw row per Ask AI conversation or one row per prompt run, depending on the required reporting grain. Do not use only account ID, date, or prompt text as a deduplication key because a conversation can contain multiple runs. A proposed key might combine account ID, conversation ID, and a prompt-run identifier, but the selected system must expose those values. No public API, webhook, or export schema for retrieving Ask AI sessions was verified.

See HighLevel’s Ask AI multi-threading guide.

Set shared rollout checks once, then test each feature’s distinct failure

Current help articles may postdate the announcement. Rollout, permissions, plan, location, and regional eligibility can vary, so confirm availability and current instructions in the account before training staff. The official sources reviewed here document user-facing workflows and constraints, not a general-purpose API or export for all five features.

Keep the controls specific to the feature. Outbound calling needs consent and suppression checks. Transfers need a reachable staff destination and escalation route. Wallet delivery needs ticket-attendee matching. Voice AI actions need confirmed action results. Ask AI outputs need review before consequential use. Record failures in the operational system the team already uses rather than assuming a new HighLevel export exists.

External ledgers, JSON contracts, idempotency keys, database uniqueness constraints, and CRM write-back designs are implementation recommendations only. Use them only after confirming that the selected integration exposes the identifiers required for the declared row grain. Separate raw observations, such as one call attempt or one wallet delivery attempt, from reported summaries and CRM contact or deal events.

  • Trigger: What starts the process, and what exact record or user action is in scope?
  • Input and owner: Which required values must be present, and who owns the source of truth?
  • Validation: What deterministic checks run before a call, write, transfer, or message?
  • Success evidence: What result proves completion rather than merely a request, spoken claim, or generated draft?
  • Exception route: Who handles missing data, a failed action, an unavailable staff member, or a ticket mismatch?
  • Measurement: What measure is reviewed at the correct grain, such as confirmed booking actions, eligible call attempts, successful staff handoffs, correct ticket scans, or reviewed Ask AI conversations?

Deploy only when the trigger, input, owner, validation, destination, and exception path are understood. For help planning account configuration, see GoHighLevel setup and implementation.