Changelog

In-call SMS and WhatsApp messaging

Voice agents can now send SMS or WhatsApp messages during a live call and feed the customer’s reply back into the conversation as a variable. This is useful for capturing details that are hard to relay by voice — one-time passcodes, email addresses, invoice IDs, links, and short menus of options.

• Two new action types, INCALL_SMS and INCALL_WHATSAPP, attachable to a voice agent under the During phase.
• Blocking or non-blocking reply handling with a configurable timeout and fallback message.
• Replies are injected into the call transcript and exposed as {sms_reply} / {whatsapp_reply} variables.
• Requires a connected Twilio account; in-call WhatsApp additionally requires a linked WhatsApp sender.

See the In-Call Messaging guide.

Website crawler for Knowledge Base

You can now crawl a website directly into your Knowledge Base by starting from a root URL. Configure crawl settings such as max pages, sitemap handling, include/exclude paths, and sync schedule.

The crawl runs asynchronously and may take a few minutes to complete, depending on site size and crawl depth.

Learn more: Crawl a Website for Knowledge Base.

Zendesk integration for Knowledge Base

You can now connect Zendesk via OAuth and import Help Center articles into your Knowledge Base.

The import is asynchronous: status moves to In Progress while resources are being indexed, then changes to Synced once complete.

Learn more: Import Knowledge Base Content from Zendesk.

New call transfer features: Preserve Caller ID, Summarization Prompt, Call Screening, and Human Detection

We’ve added four new options for live transfer actions, available both in the platform and via the API:

• Preserve Caller ID (preserve_caller_id) — Display the original caller’s phone number to the transfer target instead of the Synthflow agent’s number. Requires a number purchased through the Twilio integration inside Synthflow.
• Summarization Prompt (summarization_prompt) — Provide a custom prompt to control what information is included in the warm transfer summary briefing.
• Call Screening (screened) — Enable call screening so the transfer recipient can speak with the AI agent and accept or decline the call before being connected.
• Human Detection (human_detection) — Wait for a live human to answer before delivering the warm transfer briefing. Use with human_detection_timeout_seconds (30–1800s) to set the detection window.

For setup details, see Phone Number Transfers. For API usage, see Create Action and Update Action.

GPT-5.4, GPT-5.4 Mini, and GPT-5.4 Nano are now available

Three new OpenAI models are now available in the General Configuration AI Model selector and through the external API:

• GPT-5.4 — OpenAI’s latest flagship model.
• GPT-5.4 Mini — compact variant balancing performance and cost.
• GPT-5.4 Nano — lightweight variant optimized for speed and cost.

You can also set these models via the API when creating or updating an agent:

1
"agent": {
2
  "llm": "gpt-5.4"
3
}

Usage dashboard for subaccounts

The Usage Dashboard is now available for subaccounts, so you can review billable minute and chat usage of your subaccounts with a subscription.

To learn more about usage, see Monitoring usage.

Use SIP X-Headers in prompts, actions, and transfers

Synthflow now accepts arbitrary X-* SIP headers on inbound calls. Any custom header your SBC, PBX, or carrier attaches to the SIP INVITE is automatically extracted and made available as a variable throughout the call lifecycle — in prompts, custom actions, and outbound transfer headers.

See the full guide: Use SIP X-Headers in Prompts, Actions & Transfers.

Single Sign-On (SSO) support

Synthflow now supports Single Sign-On (SSO) for Enterprise plans. Connect your identity provider to centralize authentication and enforce access policies across your organization.

Outbound chat conversations now documented in the Platform API

The Platform API reference now includes the public outbound chat endpoint used to trigger SMS and WhatsApp conversations for chat agents.

New endpoints:

Notes:

• The endpoint uses the X-User-Api-Key header for authentication
• channel supports sms and whatsapp
• WhatsApp freeform outbound messages are limited to the 24-hour customer service window; use a Twilio content template outside that window

Clarify end_call_reasons placement on create and update assistant

The OpenAPI spec listed end_call_reasons on both the top-level request body (via Agent_model) and the nested agent object. Only agent.end_call_reasons is applied when creating or updating an agent; a top-level end_call_reasons field is ignored and does not return an error.

The duplicate top-level field was removed from the documented create/update request schema (via Agent_model). GET /assistants and GET /assistants/{model_id} responses still document an optional top-level end_call_reasons on each assistant object where the API may return it, with a note that configuration must use agent.end_call_reasons on writes. The end_call_reasons field on the Agent schema states this explicitly. The End-Call Reasons guide was updated with a full request example.

Updated references:

Unified Changelog

All platform updates are now consolidated into a single Changelog tab. You can filter entries by category using the tags: API Changes, Deprecation Notices, and Product Updates.