Akool Streaming Avatar API in 2026: Pricing, Limits and a $0.02 Option

Akool is a Palo Alto AI video company best known for face swap, video translation and avatar videos; it also hosts other labs' video models, such as Kling and Veo, and sells voice cloning. Streaming Avatar is its real-time product. Your server creates a session through the API, and Akool's avatar joins an Agora, LiveKit or TRTC room at 1080p, or up to 4K. In Dialogue mode it listens through the user's microphone and answers with Akool's own LLM and voice, optionally grounded in a knowledge base of your documents and URLs. In Retelling mode it speaks the text you send, which is how you put your own LLM behind it. Commands set the voice and language, interrupt a reply, and trigger gestures (wave, nod, thumbs up) and facial expressions. A JavaScript SDK wraps the Agora side. Everything else Akool sells, from face swap to video generation, is out of scope here: this page compares the real-time avatar only.

Akool sells credits on per-seat plans, and Streaming Avatar spends them per 10 seconds. As of 30 October 2026, API access starts on Pro Max: $59 a month for 1,200 credits, or $41.30 a month paid yearly, with larger tiers up to 9,600 credits for $449. Business is $249 a month for 6,000 credits ($174.30 paid yearly), up to 80,000 credits for $3,000. A 1080p stream costs 1.2 credits per 10 seconds on Pro Max and 1 on Business, so 7.2 or 6 credits a minute. That works out to $0.23–0.35 a minute on Pro Max and $0.16–0.25 on Business; the low end needs Business's 80,000-credit tier paid yearly. Above 1080p, up to 4K, costs twice as many credits. Enterprise API rates are custom. The credits are the same ones Akool's video, translation and face swap tools spend, and each extra admin or editor seat costs what the owner pays.

At 1,000 minutes a month of 1080p streaming you need 6,000 credits on Business, which is exactly its starting tier: $249 a month, or $174.30 paid yearly. Pro Max would need 7,200 credits, so its 9,600 tier at $449. The same 1,000 minutes cost $20 on Talking Avatar Realistic Avatars, or $60 on Talking Avatar Voice Agents, which include the AI model, speech recognition and voice. At 10,000 minutes Akool needs 60,000 credits, so Business's 80,000 tier: $3,000 a month, or $2,100 paid yearly. Realistic Avatars costs $200 and Voice Agents $600. The Realistic Avatars figure is the face for the agent you already run; Akool's includes its own speech recognition, LLM and voice. Past those volumes both sides quote: Akool's Enterprise API is custom, and our enterprise and high-volume contracts go as low as $0.005 a minute.

Akool's API pricing lists the stream's rate per 10 seconds; it doesn't say how a session that ends partway through a step is rounded. When you create a session you set its duration, up to 3,600 seconds, and Akool's docs say credits for the full duration are charged up front, with unused credits refunded after the session ends. Your plan caps that duration: 30 minutes per session on Pro Max and 60 on Business. It also caps parallel API sessions at 3 on Pro Max, 5 on Business and 10 on Enterprise. The free plan includes 5 minutes of streaming in Akool's app, but no API. Realistic Avatars bills per second with no minimum, holds nothing up front, has no concurrency tiers, runs sessions of up to 3 hours, and new accounts get $1 of free credits with full API access.

Akool's own pipeline is the default: its speech recognition, LLM and voice. To put your agent behind the avatar, the docs describe two routes. In Retelling mode your code sends the reply as text and Akool speaks it, in an Akool voice or through your own ElevenLabs or MiniMax key; speech recognition can use OpenAI Realtime. Your agent's own voice and timing don't carry over, because Akool turns the text into speech again. The second route is LiveKit Agents, where your agent's audio drives the avatar. Akool's guide installs that plugin from a branch of Akool's own copy of the LiveKit Agents repository, not from LiveKit. We found no Pipecat integration. Talking Avatar Realistic Avatars takes your agent's audio as it is, from OpenAI Realtime, LiveKit Agents, Pipecat or any TTS, and streams the face over WebRTC to your page or into your LiveKit room.

Akool sells several kinds of custom avatar. Pro Max includes 5 custom Instant Avatars made from your own video and Business 10, with one fine-tuned Studio Avatar on Business and more at $249 each; an Ultra Avatar with gestures and emotions is sold through sales. Our library is ready-made photorealistic avatars of real presenters, and a custom avatar of a real person costs $199, one time, with their consent, and works in the API and in Voice Agents. Our avatar starts speaking less than 250 ms after your agent's audio reaches it, measured on our servers; Akool publishes no latency figure.

How much changes depends on which mode you use. In Retelling mode your LLM already writes the replies, so you mostly add a voice. In Dialogue mode Akool runs the conversation, so move that first.

1. List what each Akool session uses: avatar, voice, language, mode (Dialogue or Retelling), knowledge base, and any gesture or expression commands.
2. Dialogue mode: rebuild the conversation in your own agent (OpenAI Realtime, LiveKit Agents or Pipecat) with your knowledge base as retrieval, or in Voice Agents with the key facts in the prompt and tools for lookups.
3. Retelling mode: keep your LLM and add speech output, a realtime voice model or a TTS, so your agent produces audio instead of text.
4. Create a Talking Avatar Realistic Avatars account ($1 of free credits), create an API key and pick a ready-made avatar, or order a custom one for $199, one time, with the presenter's consent.
5. Replace Akool session creation and the Agora or TRTC client with a Talking Avatar session: our player over WebRTC, or the avatar joining your LiveKit room.
6. Remove set-action gesture and expression commands from your code: our avatars show natural expressions and subtle head motion from the audio alone.
7. Run both on a slice of traffic and compare the cost per conversation before your Akool plan renews.