Skip to main content
Anam adds a real-time animated avatar to Convo Mode. While the user speaks to the Digital Twin, an embodied character appears on screen with synchronised lipsync, expressions, and natural turn-taking. Your chosen conversation model still does the thinking — Anam handles voice synthesis, animation, and the audio plane.
Anam is Pria’s avatar provider — choose it whenever you want an embodied, talking presence in Convo Mode rather than voice alone.

​
What Anam Is

Anam is a vendor that streams a photorealistic animated avatar over WebRTC, with built-in microphone capture, speech recognition, voice synthesis, and lip-sync. Pria connects to Anam as a Custom LLM — meaning your Conversation model (any provider Pria supports) generates the assistant’s words, and Anam speaks them through the avatar.
This split lets you keep all the brain power inside Pria (knowledge base, IP Vault, tools, memory, guardrails) while outsourcing the voice + video presence to a vendor optimized for it.

​
When to Use Anam

High-engagement training

Onboarding, compliance modules, role-play scenarios where an embodied character holds attention better than a text bot.

Sales & marketing demos

Greeter avatars on landing pages, interactive product walkthroughs, video-call-style sales conversations.

Accessibility

Visible mouth movements support lip-reading and comprehension for users with hearing differences.

Classroom presence

A consistent visual tutor across sessions feels more like a teaching assistant than a chat interface.
For fast Q&A, code help, or text-heavy conversations, a plain voice provider (OpenAI Realtime, Gemini Live, ElevenLabs) is usually a better fit.

​
Getting Anam Switched On

The Praxis AI team sets the avatar up for you. Choosing the character, its voice, the still image people see before it connects, its opening greeting and the Anam credentials behind it are all done by Praxis AI on your Digital Twin — there’s nothing for you to obtain from Anam first.
1

Ask for it

Contact humans@praxis-ai.com and say which Digital Twin you want an avatar on.
2

Describe the character you want

Bring the persona rather than the technical details — the kind of presence that suits your audience, the tone and accent of the voice, and the greeting you’d like it to open with. Praxis AI matches that to the Anam catalogue.
If you already have an Anam account of your own, say so — your Digital Twin can be pointed at it, and the avatar’s minutes then bill to you rather than to Praxis AI. See Cost Notes.
3

Turn the conversation on

Once the avatar exists, Enable Convo (Speech-to-Speech) under Realtime / Convo on Configuration and Integrations, and set the Speech-to-Speech Model on Personalization and AI Models to the Anam one.

​
What you can tune yourself

Once it’s running, these are yours to adjust — no request needed: Individual assistants can also override the model used for their spoken turns — see Per-Assistant Voice Model Override.

​
Connecting the LLM

Anam runs in Custom LLM mode — it does not have its own reasoning engine. Instead, each conversational turn flows like this:
  1. User speaks → the speech is transcribed in the browser.
  2. Browser sends the user’s utterance to Pria.
  3. Pria runs your configured Conversation Model (any provider — Claude, GPT, Gemini, Grok, etc.) with full access to RAG, IP Vault, tools, and conversation history.
  4. Pria streams the assistant’s reply back to the browser.
  5. Anam speaks the reply through the avatar with synchronised lipsync.
This means you can change the conversation model independently of the avatar — switch to a faster model for snappy demos, or a larger model for deeper tutoring, without changing the avatar configuration.
The Conversation Model is configured per Digital Twin in the AI Models section. The avatar uses whatever you’ve selected there.

​
Testing the Avatar

1

Open Convo Mode

From any Digital Twin chat view, click the Convo button (the microphone icon in the input area).
2

Wait for it to connect

The avatar’s still picture shows for a moment while the session connects, then the live avatar takes its place.
3

Hear the greeting

If a greeting was set up for your avatar, it speaks it once the video, the audio and your microphone are ready (after a short pause, one second by default). If you hear nothing at all, contact Praxis AI — the avatar’s setup is theirs to check.
4

Have a conversation

Speak naturally — the avatar should respond with synchronised lipsync. Tool calls (search, IP Vault lookups) happen invisibly server-side; the avatar just speaks the answer.

​
Per-Assistant Voice Model Override

The avatar and voice are configured once per Digital Twin, but each assistant can override the model used for its voice turns. In the assistant’s edit panel, set STS / Voice Conversation Model (and, if you like, STS / Voice Reasoning Effort) — it applies only when that assistant runs in a voice conversation through ElevenLabs, Anam, Cartesia or GPT-Live. If unset, voice turns use the assistant’s normal conversation model, then the Digital Twin’s voice/default model. This lets a “Sales Onboarding” assistant answer with a fast, snappy model while a “Research Tutor” assistant uses a deeper model — all through the same avatar.

​
Cost Notes

There are two bills to keep straight, and they meter different things — and the picture changes depending on whether the avatar is talking with one person in Convo Mode or has been sent into a meeting. In a one-to-one Convo Mode session: Pria credits meter the conversation. An avatar session is charged on the tokens of the spoken exchange, totalled across the session and deducted when the session ends — the same way a voice-only Convo Mode session is charged. The avatar adds no per-minute line item to the credit balance. If your Digital Twin runs on your own Anam account (arranged with Praxis AI — see Getting Anam Switched On), Anam separately bills your Anam account per minute of active avatar time and you’ll see that usage in your Anam dashboard — on top of the Pria credits the conversation consumes. On Praxis-managed Anam access there is no second bill; the avatar time is already covered by your Praxis plan. In a meeting (the avatar sent into a Zoom, Google Meet, or Teams call as the Meeting Agent), billing works differently: Pria meters the avatar, the meeting bot, and live transcription per minute of the time the bot is in the call (at one and a half times the providers’ cost), plus the twin’s spoken turns under the normal per-turn rule. If the avatar runs on your own Anam account, Anam bills your account directly for the avatar’s minutes and that portion drops out of your Pria credits — but the meeting bot and transcription are always Praxis-managed, so those per-minute charges bill in credits regardless of whose Anam account the avatar uses. Strategies to keep both down:
  • Use Anam only for the Digital Twins where the avatar matters — sales, training, accessibility. Use plain voice (Gemini Live, OpenAI Realtime) for fast Q&A Twins.
  • Ask for a clear greeting so users immediately know they’re in a voice/avatar session and aren’t surprised by the meter running.
  • Use a fast voice-turn model — a dedicated Anam conversation model keeps spoken turns short and inexpensive without changing text chat.
For Praxis-managed Anam plan details, contact humans@praxis-ai.com. For how the conversation itself is metered, see Credits.

​
Troubleshooting

  • Check Enable Convo (Speech-to-Speech) is on and the Speech-to-Speech Model is the Anam one.
  • If the avatar runs on your own Anam account, check that account still has credit — an empty Anam balance stops the session before it starts.
  • Otherwise contact humans@praxis-ai.com: the avatar and its credentials are set up by Praxis AI, so that’s where a broken setup gets fixed.
  • Usually a network jitter issue. Have the user try a wired connection or a less-congested Wi-Fi.
  • If it happens for everyone, ask Praxis AI about a different voice — some voices keep tighter lip-sync than others.
  • Check that the user’s browser allows microphone access (browser permission, OS permission, headset selected).
  • On corporate networks, WebRTC traffic may be filtered. Test from a personal network to isolate.
  • The model behind the spoken turns may be misconfigured. Check Anam Response Model — or, if it’s on the default, the twin’s Conversation Model — and any STS / Voice Conversation Model on the assistant in use.
  • Look at the chat history view after the session — if no assistant messages appear, the LLM is failing silently.
  • Confirm the browser allows video playback (autoplay policies can block the avatar stream).
  • Try the same account in another browser — if the avatar shows there, it’s the first browser’s settings, not your Digital Twin.