Tavus adds Agora support; Sume has no real-time transport to plug in

Tavus's changelog lists "Official support added for Agora, a real-time conversational WebRTC framework" on October 8, and Sume has no equivalent: there is no Agora, WebRTC, LiveKit or other real-time transport for Sume avatars. A Sume avatar request is an HTTP job that returns a finished MP4 you fetch by URL.

The Tavus facts come from the Tavus changelog, read 2026-10-11. The Sume facts come from Generate avatar video and Jobs and results.

What the Tavus changelog shows

Near the top of the changelog, in the week of October 5 to 9, there are four relevant entries. On October 9 the PAL Maker gained internet access for the PALs it creates. On October 8 Agora support was added. On October 7 Phoenix-4.5 improved facial emotions and reduced micro twitches for video-based faces. On October 5 Pipecat and LiveKit pipelines got lower audio-to-frame latency.

Three of those four are about real-time delivery. That is the direction of the real-time avatar market this month, and none of it carries over to a render-by-request API.

Notice what is absent: nothing in these entries is about the look of a recorded video. Even the Phoenix-4.5 note concerns facial emotion and twitch reduction in a streaming face model. If you came to the avatar space for finished video, the live-agent news mostly does not apply to you, and that is a good reason to keep two separate checklists.

How the two designs differ

The comparison is about the delivery model, not quality.

What a Sume job gives you instead

You submit a request with an avatar_handle and a script, receive a job id, and get a webhook or poll GET /v1/jobs/:id/status until the job is complete, then read the result. Completed results can include media.sume.com video artifacts plus preview fields. There is no socket to hold open, so a serverless function or a queue worker can handle the whole flow.

That shape suits personalized outreach, onboarding clips and recurring updates. It does not suit an assistant that answers a visitor mid-sentence. For that case the honest advice is to use a real-time vendor such as Tavus and keep Sume for the pre-recorded parts, for example a welcome clip that plays before the live session opens.

Pricing differs in kind as well. A Sume avatar job is priced per second of finished video before you submit, with published rates by tier. A live session is priced by the vendor's own session or minute model, which this post does not quote because I did not read a Tavus pricing page for it.

A hybrid that works

Many teams that evaluate live avatars end up with both: a rendered opener that always plays, then a live agent for questions. The opener is an avatar clip you can approve in advance, with a disclosure line in the caption. The live part is whatever transport your real-time vendor supports, now including Agora for Tavus.

If you want to try the rendered half, create the avatar once, write a 20-second welcome script, and render it on the standard tier. Reuse the handle for every segment of your site, and keep the clip short so it never delays the live session.