Chat Widget

You can now embed a Retell Chat agent directly into your website or app — enabling users to interact with your AI agents via a modern, lightweight text UI, without any setup overhead.

How to set it up:

1. Copy and paste the embed script into your website

2. Replace the publicKey, agentId, and version

3. Customize widget name

Copy paste the embed code to add your chat agent— no backend work required.

See Docs

Public Keys for Chat Widget

Unlike API keys, which must remain private, Public Keys are designed for frontend use and are required for authorizing access to the new chat widget.

Manage your keys here:

Dashboard → Keys → Public Keys

See Docs

Custom Function Enhancement

You can now use Custom Functions to:

• Make GET, POST, PUT, PATCH, and DELETE requests
• Add custom headers and query parameters
• Better UI when entering parameters

See Docs

Fast Tier For OpenAI models

The former “High Priority” model option is now Fast Tier, with upgraded performance and a reduced price:

• Predictable latency

• Uncapped scale

• Higher reliability

• New pricing: 1.5x model cost (down from 2x)

Learn More from OpenAI

Improved Outbound Diagnostics: New Telephony Error Codes

To give you better visibility into failed calls, we’ve expanded the list of disconnection reasons and restructured the call status reporting for outbound dials.

New disconnection_reason values:

• invalid_destination: Number is malformed or unreachable
• telephony_provider_permission_denied: SIP trunk auth issue
• telephony_provider_unavailable: Provider is offline or failing
• sip_routing_error: SIP loop or routing misconfig
• marked_as_spam: Blocked by spam filter
• user_declined: Callee manually declined the call

🗓 Effective Date: June 30, 2025

Ensure your integration handles these new values properly.

See Docs

Medical Transcription Mode

We’ve launched a medical vocabulary setting for English voice agents. When enabled, transcription will favor terminology commonly used in clinical and healthcare scenarios — improving accuracy for patient-facing use cases.

Currently available for English agents only.

Other Improvement

1. Custom function reply response limit has been increased to 15,000 characters, giving you more flexibility when returning detailed responses.
   
2. LLM prompt token limit is now set to 32,768 tokens, enabling more complex and context-rich conversations.
   
3. For importing custom telephony numbers, we removed the limit of e.164 format so you can import PBX stuff / pseudo number.‍
4. Call History table & CSV exports now include Agent ID and Version→ Enable these via Customize Fields
5. You can now write a prompt for voicemail.
6. Added a few new languages: Dutch (Belgium), Catalan and Thai.

‍