June 1, 2026

Agents can now search the web

Agents can now search the web during conversations. Web search is enabled by default — set useWebSearch to false when creating or updating an agent if you want the agent to rely only on its provided knowledge.

The field is available on agent creation, agent updates, channel overrides, and connection requests.

See Agent Configuration — Web search for details.

Create agents from real people

You can now create digital twins — companions based on a real person's likeness and voice. Unlike custom companions, a digital twin uses the person's actual voice recording and appearance to produce an avatar that looks and sounds like them.

New endpoint

• POST /public/digital-twins — create a digital twin from a voice recording, reference image or video, and identity details

Key fields

• voiceUrl (required) — a recording of the person's voice, used to clone it for the avatar
• externalClientId (required) — ties the digital twin to a real individual in your system
• videoUrl (optional) — reference video for higher quality lip sync and expressions

Once created, a digital twin returns a companion ID that works the same way as any other companion — attach it to an agent and deploy to any channel.

See the Digital Twins guide for details.

Arabic-first LLM with HUMAIN's Allam

The Napster API now supports HUMAIN as an LLM provider alongside Azure OpenAI. HUMAIN is powered by Allam, an Arabic-first foundation model built by HUMAIN in Saudi Arabia with strong dialect coverage and cultural alignment — best for Arabic-language agents.

HUMAIN is not enabled by default. Contact us to request access.

To use HUMAIN, select it as the Provider when creating a new API key.

Higher quality avatars with video input

You can now choose between two avatar models when creating a custom companion or digital twin. Both produce real-time video avatars — the difference is the reference input you provide and the resulting quality.

v1 (default)

Uses a still image (pictureUrl) as input. Fast generation, good quality, and strong photorealism.

v2

Uses a video of the person speaking (videoUrl) as input. Highest quality with the best lip sync accuracy, natural facial expressions, and better results with facial hair. Generation can take up to several hours.

New fields

• version — set to "v1" or "v2" when creating a companion. Defaults to "v1" if omitted.
• videoUrl — URL of a reference video for v2 generation.

Example

curl -X POST https://companion-api.napster.com/public/companions \
  -H "X-Api-Key: $NAPSTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "firstName": "Alex",
    "description": "A friendly customer support specialist.",
    "videoUrl": "https://example.com/videos/alex-speaking.mp4",
    "version": "v2"
  }'

See Avatar Models for the full comparison.