HighLevel’s September 25, 2026 Release Radar announced five updates for the week of September 21 to 25: the Managed Agents Public API, AI Studio migration to SSR, AI Studio’s rollout beyond Labs, Calendar AI availability troubleshooting, and mobile reschedule-link generation. The useful question is not simply what was announced. It is which operating path is documented today, what must be verified in the target account, and where a human or deterministic application rule remains responsible.
This guide separates announcement claims from current HighLevel support and API documentation. It turns each update into an operating pattern with a trigger, a defined AI or product task, validation controls, a destination, and an exception path. Product interfaces, eligibility, permissions, API schemas, and support pages can change, so confirm the relevant documentation and account before deployment.
An external agent call, a project migration, an availability investigation, and a mobile share action are different workflows. Treating them as interchangeable feature launches is how teams lose track of ownership, duplicate requests, or report an action as complete when only an intermediate step occurred.
What the September 21 to 25 HighLevel Release Radar announced
HighLevel’s September 21 to 25 Release Radar covers five operating areas:
- Managed Agents: external applications can invoke published agents through an API.
- AI Studio migration: eligible React/Vite projects can be migrated into separate TanStack Start drafts for SSR.
- AI Studio access: the announcement describes a rollout beyond Labs and a permission workflow.
- Calendar AI: users can investigate why availability is blocked through conversational troubleshooting.
- Mobile reschedule links: staff can generate a link from a supported appointment and open the device share sheet.
The Release Radar establishes what HighLevel announced. The current product and API documentation below establishes the safer operating detail. It does not establish that every permission screen, appointment type, account, or API response is identical across locations.
Managed Agents API: invoke a published agent, not a draft
HighLevel documents external invocation for a Managed Agent configured with the Invoked via API trigger and published. The documented execution endpoint is POST /agent-studio/managed-agents/{managedAgentId}/execute. Requests use API version v3, a bearer token with the managed-agents.write scope, and an Idempotency-Key no longer than 128 characters.
The Managed Agents deployment guide covers setup, publication, tokens, sessions, and activity. The execution endpoint reference covers request-level requirements. Use the current endpoint schema rather than treating a generated cURL, Python, or JavaScript sample as a complete production integration.
- Configure the agent’s Invoked via API trigger, test it, and publish the intended version.
- Create a Private Integration token for the intended sub-account with the required scope. Keep it on the server, never in browser code, screenshots, or a repository.
- Authenticate the caller, select the tenant, verify that the requested agent is allowlisted and published, and validate required input before calling HighLevel.
- Omit
sessionIdon the first turn. Store the returned session ID against the application’s authenticated user or conversation record, then send it on later turns that should continue the same context. - Validate the result before returning it to the application or writing to a CRM record. Route consequential changes through deterministic business rules and any required human approval.
Authentication, tenant selection, permissions, required fields, consent, and duplicate detection belong in application logic. The agent should perform the bounded language task configured for it. For a retry of the same logical turn, reuse the same idempotency key. For HTTP 429 responses, follow the documented retryable and Retry-After fields rather than retrying blindly. A database uniqueness constraint or atomic insert or upsert is required to protect against concurrent duplicates.
For an external caller, the published agent is the API release artifact. Publishing does not replace the application’s own release checks, validation, monitoring, or approval controls.
A proposed application record
The following is an illustrative application model, not a HighLevel schema. Its grain is one logical API turn. A conversation can contain many turns, so do not use only the session ID as the unique key.
{
"tenant_id": "location-123",
"managed_agent_id": "published-agent-id",
"session_id": "returned-session-id-or-null",
"logical_turn_id": "conversation-42-turn-3",
"idempotency_key": "turn-2026-11-22-00042",
"request_hash": "application-computed-hash",
"status": "pending"
}
For retained request records, enforce uniqueness on (tenant_id, idempotency_key) and, where appropriate, (tenant_id, logical_turn_id). Those are proposed application controls. A lookup followed by create is not race-safe when two workers process the same request. Use a database-enforced constraint with an atomic insert or transactional upsert. Map the session to the application’s business conversation, and reject a continuation request if its tenant or user context does not match.
API calls execute the published version. Draft edits do not change external behavior until publication. Unpublishing can cause invocation to stop and return a 404. A 404 may also indicate an unknown or cross-location agent, so record the tenant and agent context when troubleshooting.
When the work shifts from API mechanics to defining a bounded business role, AI agent implementation is the relevant ConsultEvo service reference.
AI Studio migration: validate the SSR copy as a separate project
HighLevel documents migration from an eligible React/Vite AI Studio project into a separate TanStack Start draft. This is not an in-place conversion of the live original. The original project, version history, published site, and connected domains remain unchanged. Domains are not automatically transferred to the migrated copy.
New AI Studio projects use TanStack Start and support SSR, Server Functions, REST APIs, and secure Secrets. Existing React/Vite projects require the separate migration workflow. The migration guide describes the product procedure and the need to review the result. The AI Studio SSR and Server Functions guide describes the current architecture for new projects.
Test rendered pages, forms, calendars, integrations, secrets, workflow associations, domain configuration, and any workflow filters that reference the original project. AI-assisted conversion may require corrections. Keep the original live until the owner approves the candidate and any publication or domain cutover.
AI Studio access: distinguish the announcement from verified account setup
The September 25 Release Radar announced that AI Studio was rolling out beyond Labs and described a Home dashboard permission workflow for agency owners and administrators. The current reviewed technical documentation does not independently confirm the precise dashboard procedure or universal availability. An earlier AI Studio overview still describes Labs-based enablement, so treat the announcement as an account-level availability claim to verify, not as a promise of identical access everywhere.
An administrator should inspect the target account’s current interface and role permissions, grant access only to intended users if the available interface supports it, and confirm access before assigning project work. HighLevel says access should become available shortly after permissions are granted, but the reviewed technical documentation does not independently establish a universal activation time. Access administration is separate from migration: giving a user AI Studio access does not migrate an existing React/Vite project.
Calendar AI: investigate the reason before changing availability
Calendar AI is a conversational aid for investigating why time is unavailable. Current documentation supports questions about a particular slot, day, calendar, team member, or a team member serving multiple calendars. Include the date, time, calendar, and time zone for a precise slot question. The system may request clarification when calendar or team-member context is ambiguous.
Open Calendars, launch Calendar AI, ask a scoped question, answer clarification prompts, and review the explanation. Documented conditions include working hours, scheduling notice, buffer time, booking limits, blocked time, external-calendar conflicts, unavailable resources, and other restrictions. If Calendar AI cannot provide a reliable supported reason, use Troubleshooting View instead of inferring one.
Use Calendar AI to narrow the investigation, then have the calendar owner verify the schedule, resource, conflict, and permission settings before changing them. A whole-day response is a summary of repeated conditions across time segments, not a formal export of every raw slot.
The Calendar AI troubleshooting guide describes the supported scope and fallback. If your application logs investigations, use one record per user query or troubleshooting session. Store slot-level observations separately only when that evidence is intentionally required. Do not use calendar ID and date alone as an investigation key because multiple users can ask multiple questions about the same day.
Mobile reschedule links: check appointment eligibility before sharing
Current mobile documentation limits the verified capability to supported Service Calendar v2 appointments. It does not independently verify the Release Radar’s broader wording about all Meetings calendars or every appointment status.
Open the supported appointment in the mobile app, tap the three-dot menu, choose Generate reschedule link, review the link, and tap Share. The link is copied and the device’s native share sheet opens. The current support page specifies HighLevel mobile app version 4.10.0 (810) or later.
If HighLevel cannot build the link, the interface displays N/A and disables Share. Once the share sheet opens, the staff member must choose an installed messaging or email application and complete the send. The reviewed documentation does not establish delivery confirmation through SMS, WhatsApp, email, a public API, or an external automation.
Separate appointment eligibility, link generated, share sheet opened, and message sent. Before telling a customer the link was sent, confirm the supported appointment, successful generation, and completion of the send in the selected application. Treat N/A with disabled Share as a generation failure.
The Service v2 mobile appointment guide documents the action path, app requirement, native share behavior, and failure state. If auditability matters, store one application record per link-generation event or use an explicit appointment-level upsert for the latest usable link. Do not claim that repeated generation creates a new business event unless your application defines and records that event.
Choose the operating pattern that matches the update
| Trigger | AI or feature job | Validation and action | Fallback owner |
|---|---|---|---|
| Authenticated application request | Published Managed Agent performs a bounded conversational task. | Backend verifies tenant, allowlist, publication, idempotency, and result before returning or writing to the CRM. | Integration owner handles API errors; business owner reviews consequential actions. |
| Eligible React/Vite project | AI Studio creates a separate TanStack Start draft. | Project owner tests behavior, dependencies, secrets, and domains before publication or cutover. | Web operations and integration owners handle defects. |
| Availability question in Calendars | Calendar AI explains supported blocking conditions or asks for clarification. | Calendar owner verifies the underlying configuration and decides whether a change is appropriate. | Scheduling administrator uses Troubleshooting View when unresolved. |
| Supported Service Calendar v2 appointment | Mobile app generates a link and opens the native share sheet. | Staff confirms eligibility, generates the link, and completes the send in the selected application. | Appointment owner investigates N/A or disabled Share. |
AI Studio access belongs in the administrative checks for a rollout plan. It is not an external integration and does not substitute for migration testing.
A practical adoption sequence
Start with the operational problem and its owner, not with enabling every feature at once. Select one workflow, document its current path, test a normal case and a meaningful exception, establish a baseline, and expand only when the result is useful.
- Managed Agents: establish a published-agent release gate, server-side token handling, tenant and session mapping, idempotency, and a review path before allowing CRM writes. Use CRM systems and operational data flows when defining the system-of-record boundary.
- AI Studio migration: inventory the live project, preserve the original fallback, test the separate draft, and require owner approval before publication or domain cutover.
- AI Studio access: verify the current account interface and intended user permissions before assigning project work.
- Calendar AI: assign a calendar owner to verify the explanation and approve any configuration change.
- Mobile links: define success as a generated link for an eligible appointment that staff actually shares, not merely a tap on Share.
- Name the workflow owner and the system that holds the final business record.
- Verify the documented eligibility path in the target account, project, calendar, appointment, or app.
- Test one normal case and one real exception, such as an unpublished agent, incomplete migration, unclear availability cause, or
N/Alink. - Define the destination event separately from intermediate observations, such as a returned session, generated draft, diagnostic explanation, or opened share sheet.
- Choose one measurable signal, such as duplicate API turns, migration defects found before publication, unresolved availability investigations, or link-generation failures.
For broader account planning, GoHighLevel setup and implementation is a relevant starting point. The practical decision is simple: choose one owned workflow, verify its exception path, and expand only when the measured operating result justifies it.
