shapeshift your agents in realtime
Faces are compiled soul documents.
They give agents programmable personas
as nuanced as real humans using fewer tokens.
More personalization, lower cost.
How it works
Faces use a minimum formal language to express persona more richly than natural language with a fraction of the tokens.
Soul Compiler
Use with any model
Our inference API routes through to all major providers, just pass in facename@model as your model to invoke a face.
Works with Anthropic, OpenAI, xAI, Minimax, Deepseek, Qwen, and more.
Example Use Cases
Use your imagination, or task your agent to do so for you. Here's a few obvious ideas:
Train a face on my existing YouTube videos and then create three sample transcripts for my next video on dating advice.
meI'll use yt-dlp to download the videos, send them to AssemblyAI for transcription, and then send the transcripts to manyface to train your face. Gimme five mins. Brb.
my agentClone your voice
Pricing
Create your faces on our platform and then use them with any other API for a 5% markup.
Starter
→ Up to 3 faces
→ 5k token compiler credits
Free
Standard
→ Up to 100 faces
→ 100k/mo token compiler credits
$5/month
Pro
→ Up to 10,000 faces
→ 1M/mo token compiler credits
$49/month
All plans come with $5 API credit grant for immediate testing.