Words matter.

Faces distills documents, interviews, and videos into cognitive primitives: the minimal code a mind uses to determine which word comes next.Use with any LLM for specialized agents and simplified teams.

FACES

Quickstart

faces-skill

Copy-paste this to your agent

Install Faces skill:
1. git clone --single-branch --depth 1 https://github.com/facessh/faces-skill.git ~/.claude/skills/faces-skill && cd ~/.claude/skills/faces-skill && ./setup
2. list the available skills: /faces, /face, /team, /manyface
3. set me up to use faces: ask if I have an account, else help me get one

Make a Face

Feed the soul compiler documents, conversations, and videos. It extracts cognitive primitives, the minimal instruction set that defines how a person uses language. The result is more complex, nuanced, and coherent than any markdown file can produce.A single Face replaces thousands of tokens of system prompts with a compressed cognitive architecture that produces richer output for less.

Soul Compiler

The soul compiler breaks down plaintext source documents into the building blocks of mind.
Soul Compiler

Compose Faces

Faces are made of embedded primitives. Use algebra to compose faces programmably. Union two minds. Intersect their common ground. Subtract a limitation. A/B test your A|B-C face.

Garry Tan | my-domain-expert - syncophant
Boolean algebra for minds. Union, intersection, difference, symmetric diff. Every composition is a real, chattable Face.

Face Math

Use Boolean logic to create composite faces with the features you want.
Face Math

Face Legos

Use faces like Legos to compose context for targeted conversations.
Face Legos

Use Anywhere

Route any inference request through the Faces API. Set your model to face-alias@model and we proxy to the provider with the Face applied. Works with Claude, ChatGPT, Grok, Gemini, Deepseek, and any open source models offered by Fireworks and Venice.

elon-musk@claude-sonnet-4-6
Your faces are yours. They don't live inside ChatGPT or Claude. They travel with you across models, agents, and platforms. One API. Any model. Many faces.

Face Namespace

Works with any existing SDK, framework, library, package, or application that uses LLMs via API.
Face Namespace

Inference Router

The Faces API adds your faces and proxies your request to the underlying LLM provider.
Inference Router

Cast with /face

Type /face "Albert Einstein" and the agent does the rest. It searches for source material — YouTube talks, Wikipedia, interviews, writings — and sketches a FACE.md recipe you can review and edit before anything compiles. When you're satisfied, it compiles the sources and gives you a /face-albert-einstein slash command you can call from anywhere.One command creates a mind. The agent does the research. You do the review.

HR for Zero-Human Companies

Create your ideal top brass and let them make their ideal workforce.
HR for Zero-Human Companies

Guru Yoga for Bots

Make faces that can act as mentors and guides to other faces.
Guru Yoga for Bots

Orchestrate Teams

Need a dream team? Type /faceteam and define a task. The Faces skill will help build an optimal team of faces and a collaboration protocol defined in a TEAM.md file with a mermaid flowchart that is human readable and easy for you or your agent to edit.Replace n8n, Paperclip, CrewAI, and other multi-agent orchestration tools with a single markdown file.

HR for Zero-Human Companies

Create your ideal top brass and let them make their ideal workforce.
HR for Zero-Human Companies

Guru Yoga for Bots

Make faces that can act as mentors and guides to other faces.
Guru Yoga for Bots

/manyface any Skill

Got an existing AI workflow that sounds like the same helpful assistant at every step? Type /manyface and point it at any skill. It decomposes the workflow into roles, casts the dream team and outputs a new version where every step is run by the ideal face for the job.A design review run by Jony Ive. Werner Herzog editing your YouTube short. A heterogeneous writers room that debates jokes to consensus. Same skill, completely different output.

Face Space

Use latent space clustering to analyze and manage your faces.
Face Space

Face-to-Face

Use cosine similarity to identify and test affinity between faces.
Face Space

What People Build

The cognitive primitives of Faces are upstream of everything else that determines how an LLM responds. This is the machine code of agentic engineering.Ask your agent to brainstorm uses for Faces based on your unique goals.

Train a face on my existing YouTube videos and then create three sample transcripts for my next video on dating advice.

me

I'll use the video transcript downloader skill then use the text to create your face. Should take about five minutes, hold on.

my agent

Clone your social media voice

Build me a board of advisors for my startup. I want Garry Tan for product, a skeptical CFO type, and someone who thinks like Octavia Butler about long-term consequences.

me

Garry's already in your catalog from last week (47k compiled tokens). I'll cast the CFO and the speculative thinker, then set up a round-robin team. Want them to deliberate or vote?

my agent

Advisory board for your company

Manyface my code review skill. I want a security-focused critic, an architecture thinker, and a pragmatic 'will this ship?' voice. Round robin until they agree.

me

I'll read the skill, cast three faces, set up the team protocol, and output a manyfaced version. Give me two minutes.

my agent

Manyface your existing skills

Build faces for each of our target audiences and then simulate a focus group for the three strategy directions we talked about this morning

me

Sounds good. I'll scrape some Instagram accounts from influencers who best represent our demographics as a starting point.

my agent

Simulate a target audience

Run a series of virtual brainstorms with the faces for each of our product team members over the long weekend and summarize the top three ideas.

me

Sure! I'll separate them into random breakout groups and simulate a hackathon. I'll judge working demos and you can try out the best prototypes on Tuesday.

my agent

Emulate your existing workforce

Build me a marketing guru face. Use Tony Robbins videos - I think there's an early one of him giving a seminar - and that Wolf of All Streets guy.

me

Found it, it's a young Tony drawing on a flip board. And you're talking about Scott Melker. I'll filter his YouTube for most popular.

my agent

Distill the expertise of industry leaders

Pricing

Use Faces with underlying LLMs for a 5% markup on the provider's per-token cost. Subscribers receive monthy compiler credits, API only users are billed per-token to compile.Both plans include unlimited faces.

Connect (OAuth)

Connect (OAuth)

→ 100k monthly compiler credits
→ Use with your OpenAI plan (via Codex OAuth)

$17/month
API Only

API Only

→ pay to compile and use
→ min $5 balance to start

5% fee
on provider rate