Changelog

September was a busy month at Synthflow: you can now build agents from inside ChatGPT, have your agent email a summary once the call ends, see exactly how every transfer ended, and move your workspace to the new, faster Synthflow app. Let's dive in.

If you missed August, it brought Campaigns, the Advanced Synthflow LLM, Business Hours, and analytics across subaccounts.

Official ChatGPT Plugin

Synthflow's official ChatGPT plugin is now live. Build and manage voice agents from a conversation in ChatGPT, without leaving the chat. It suits teams that already draft prompts and plan call flows in ChatGPT and want to turn that work into a running agent in the same place. The setup guide walks you through connecting your workspace.

Post-Call Emails

Your agent can now send an email once the call ends. Attach a Post-Call Email action to any agent and pick who should receive it. Use it to put a call summary in front of a sales rep, alert a support lead, or keep a manager in the loop, without anyone opening the dashboard.

The New Synthflow App

The new Synthflow app runs at app.synthflow.ai, with faster page loads and more responsive navigation. Turn it on for your workspace from Early Access. The guide covers what changes and the current known issues, so you can decide when to switch your team over.

Improvements

• Simulations in the agent editor. Create and run test suites without leaving the agent editor. Change a prompt, rerun the suite, and see what passed or failed right next to the agent.

• Prompt Reuse metric. Prompt Reuse in Analytics shows how much of an agent's prompt the AI model reused instead of reading it again. The more an agent reuses, the lower its latency in the call, which helps explain why some agents respond faster than others.

• 200 voices for Synthflow TTS. Synthflow TTS now offers 200 voices instead of 12. Filter the Voice selector by accent, gender, or language, and preview a voice before you assign it. Voices already on your agents are unchanged.

• Synthflow TTS for multilingual agents. Agents set to Multilingual can now use Synthflow TTS as their voice model or fallback. One voice covers the whole conversation, including when the caller switches language mid-call.

• Custom action response time. Call details now shows the response time of every custom action. Open the Actions tab to find the slow endpoint behind a long silence.

• Call quality on every call. The Telephony tab now reports call quality for every call, so you can rule out the network without pulling a PCAP.

• Auto-overage for subaccounts. Parent accounts now control whether a subaccount can call past its included minutes. Turn on Auto-overage and the extra minutes are billed to your account at your per-minute rate.

API Changes

• Time to human on warm transfers. Connected warm transfers now include human_connected_at, the moment the caller reached a person (transfer timestamps).

• New transfer outcome states and failure reasons. Transfers can now report phone menu, voicemail, and busy outcomes, and failed transfers return a specific failure reason. Find every state and reason in the transfer outcome states reference.

• Every transfer attempt in the post-call webhook. The action execution timeline lists each attempt and the number it dialed.

• Call quality in the API. The call quality measurements from the Telephony tab are also returned by the Calls endpoints.

Deprecations & Migrations

All Voice Widgets now use the new design. As of September 16, 2026, every remaining Voice Widget runs on the upgraded design and faster real-time backend. No action was required: the embed already on your site keeps working. The Voice Widget migration guide explains what changed.

August was a busy month at Synthflow: outbound campaigns can now be launched and managed through the API, a new Advanced Synthflow LLM takes on your most complex prompts, agents can keep office hours, and analytics now span every subaccount you manage. Let's dive in.

Campaigns

Launch and manage outbound campaigns through the API. Hand Synthflow a list of recipients, such as reminder waves, follow-ups, or reactivation lists, and it works through the calls for you while you follow progress in the portal under Campaigns. Available on Enterprise plans; contact support to switch it on.

Business Hours

Your agents can now keep office hours. Set weekly business hours per agent, evaluated in that agent's own timezone, and choose what happens to an inbound call that arrives outside them: decline it, transfer it elsewhere, or run a custom action. Outbound calls placed while the agent is closed are declined automatically.

This matters most when an agent fronts a team that isn't 24/7: after-hours callers reach a defined path instead of a conversation nobody is around to finish.

Advanced Synthflow LLM

Meet Advanced Synthflow LLM, the strongest reasoning model in the Synthflow family. It is built for prompts that have grown real complexity, with deep branching, layered conditions, and long instruction sets, and it holds its accuracy at scale where a lighter model starts to drift.

Synthflow LLM stays the default, and the right choice for most agents. Reach for Advanced when your prompt has outgrown it.

Analytics Across Subaccounts

The analytics dashboard now spans every subaccount you manage. Set the All accounts filter to compare call activity across subaccounts at a glance, then select any row to drop into that subaccount's full dashboard in place, with no account switching.

For resellers and agencies, it is the view that surfaces an outlier before the customer calls you about it.

8x8 Contact Center Integration

If your contact center runs on 8x8, you can now hand inbound calls in the US and EU to a Synthflow agent without rebuilding the routing you already have. Import an 8x8 number in Synthflow to get a 12-digit routing identifier, then point an 8x8 Forward to External IVR flow at it. The agent picks up, and everything upstream stays exactly where it is.

Three places teams start: AI-first triage on the main line, after-hours and holiday coverage, and taking tier-1 volume off the queue before it reaches a person.

Filter Call Logs by Phone Number

Call logs can now be filtered by the customer's number, for when a number is all you have to go on. The filter matches calls both to and from it, and combines with the existing date, agent, duration, type, and status filters.

Current Date and Time System Variable

The new {agent_current_time} system variable returns the current time in the agent's timezone as ISO 8601 with a UTC offset, and works in prompts and custom actions alike.

Pin Aurora Conversations

Pin a thread in Aurora and it stays at the top of the list, so the work you keep returning to is always in reach.

API Changes

• Subaccounts API: reset minutes used. Set minutes_used to 0 on PUT /subaccounts/:subaccount_id and the customer's usage cycle starts fresh, so you can run their subscription on your own billing schedule rather than ours.

• Phone Numbers API: import Five9 and RingCX numbers. Five9 and RingCX numbers can now be imported through POST /custom-numbers. Synthflow sets their SIP domain automatically, joining Zoom as auto-configured providers on that endpoint.

• MCP server: update_contact and delete_contact. The Synthflow MCP server now covers the full contact lifecycle. The two new tools sit alongside the existing create and lookup ones, so you can manage contact records end to end from Claude, Claude Code, or Cowork.

Deprecations & Migrations

• Reminder: Stripe-based reselling ends September 15, 2026. Custom pricing plans are being discontinued: from that date Synthflow stops billing subaccount minutes through Stripe on your behalf, and subscriptions created through them stop working. If you resell minutes this way, the reselling migration guide covers moving your customers across. Individual subscriptions are unaffected.

• Non-enterprise plans now run on Synthflow LLM. Third-party models, including the GPT family, are no longer available outside Enterprise. Existing agents were migrated to Synthflow LLM automatically, so no action is needed in the dashboard. API requests that set llm to a GPT value still return normally, but the agent stays on Synthflow LLM.

A note on speech recognition. Speech Recognition now offers a clearer choice: Faster for the lowest latency, or High Accuracy when a call depends on capturing phone numbers and order IDs exactly, at some cost in latency.

July was a landmark month for Synthflow. CMP evaluated 27 Voice AI providers and named Synthflow one of just two Pioneering companies—placing us at the leading edge of the category. That recognition came alongside major releases across Claude, Five9, live call monitoring, WebSocket media, and background audio.

Only 2 of 27 providers were named Pioneering—including Synthflow.

CMP is the research organization behind Customer Contact Week.

Here’s what shipped this month.

Official Claude Connector and Plugin

Build, run, and improve your voice agents without leaving Claude. The official Claude connector and plugin let you use supported Synthflow capabilities from Claude, Claude Code, and Cowork. On top, the plugin adds guided call and prompt reviews so making your agents better becomes a conversation.

Five9 Handoff

Available to Enterprise customers in the US and EU after activation. If your contact center runs on Five9, the Five9 Handoff action returns the call to your Five9 workflow, passing routing context as X-Headers on the SIP BYE. Your agent handles the AI portion of the call and hands control back to Five9 when a human is needed, preserving your existing routing.

Live Call Monitoring

Authorized users can listen to an in-progress call from Call Logs. Monitoring is listen-only: the monitor is not audible to the caller or agent. Ensure monitoring complies with your organization’s policies and applicable notice, consent, and privacy requirements.

WebSocket Media for Websites and Apps

Use WebSocket media to attach an authenticated media leg to a running call from a website or application. Treat the media URL and sid as temporary secrets.

Background Audio on Running Actions

Custom actions and individual MCP tools can now play background audio while they run. Callers hear something (hold music, a "one moment" message) during slower requests, instead of dead air that makes them wonder if the agent is still on the line.

Improved System Status Page

The Synthflow status page has moved to a new platform with more reliable incident reporting and status updates. The URL is unchanged and email subscriptions were migrated automatically. RSS subscribers need to resubscribe on the new status page to keep receiving updates.

Rename Aurora Conversations

You can now rename conversations in Aurora. Give each thread a clear, descriptive name so long-running work stays organized.

Sidebar Control

You can now set the sidebar to collapse, expand, or expand on hover. The control lives in the bottom-left corner of the interface.

API Changes

• Phone Numbers API: SIP UAC registration. The Phone Numbers API now supports SIP UAC registration settings when importing a number or updating an existing one.

Improvements

• Hebrew available. Hebrew joins the list of available languages that Synthflow voice agents can speak and understand.

Deprecations & Migrations

• Stripe-based reselling is ending. If you use custom pricing plans to bill customer subscriptions through Synthflow, you'll need to migrate before September 15, 2026, when the feature is discontinued. Your own Synthflow subscription is unaffected. See the reselling migration guide for the required changes and next steps.

• Global region egress IPs updated. Six new outbound IP addresses have been added for HTTP requests from the Global region: 34.73.34.138, 34.26.71.171, 34.148.30.10, 34.138.149.224, 34.139.16.79, 34.23.185.86. If your firewall allowlists inbound HTTP traffic from Synthflow, add these addresses to prevent those requests from being blocked. Full list on the Egress IP addresses page. No action is required if you don't restrict inbound HTTP traffic.

That’s everything for this month. We’ll keep sharing the product changes that make Voice AI easier to deploy, operate, and improve.