faces（faces 开发）

Faces Skill

You have access to the faces CLI. Use it to fulfill any Faces Platform request.

Always use --json when you need to extract values from command output.

Current config

!faces config:show 2>/dev/null || echo "(no config saved)"

Setup

Verify credentials: faces auth:whoami. If no credentials exist, see references/AUTH.md for registration (requires human payment step) and login.

Install (if faces command not found): npm install -g faces-cli

faces auth:* and faces keys:* require JWT. Everything else accepts JWT or API key.

Plans

Two plans: Free ($5 minimum initial spend, pay-per-token with 5% markup on all usage including compilation) and Connect ($17/month, 100k compile tokens/month, free passthrough to OpenAI Codex for users with a ChatGPT subscription). See references/AUTH.md for details.

Core workflow

1. Create a Face with basic facts and a default model: faces face:create --name "Name" --username slug --default-model gpt-5-nano --attr gender=male --attr age=34 --attr location="Portland, OR" --attr occupation="nurse practitioner"
2. Compile source material in one step: faces compile:doc slug --file document.txt

- This creates the document, runs LLM extraction with real-time chunk progress, and syncs automatically - Repeat for each source document

1. Chat through the Face: faces chat:chat slug -m "message" (auto-routes to the correct API based on model provider)
2. Compare Faces: faces face:diff or faces face:neighbors
3. Compose new Faces from boolean formulas: faces face:create --formula "a | b"

Note: compile:doc handles the full create → compile pipeline. For an already-created document, use compile:doc:make <doc_id>. Threads use a separate workflow: compile:thread:make (same fire-and-forget pattern as documents).

Boolean operators: | (union), & (intersection), - (difference), ^ (symmetric difference). Parentheses supported: (a | b) - c.

Common tasks

Create a face with attributes

When creating a Face, set basic demographic facts with --attr KEY=VALUE (repeatable). Common keys: gender, age, location, occupation, education_level, religion, ethnicity, nationality, marital_status. Unrecognized keys are silently ignored — see references/ATTRIBUTES.md for the complete list of accepted keys.

faces face:create --name "Marcus Rivera" --username marcus \
  --default-model gpt-5-nano \
  --attr gender=male --attr age=34 \
  --attr location="Portland, OR" \
  --attr occupation="nurse practitioner" \
  --attr education_level="master's degree" \
  --attr marital_status=married

You can also add or update attributes and set the default model on an existing Face:

faces face:update marcus --attr religion=Catholic --attr ethnicity="Mexican American"
faces face:update marcus --default-model claude-sonnet-4-6

Compile a document into a face

# Recommended: one-step compile with progress
faces compile:doc <face_id> --file notes.txt --label "Notes"

# Alternative: create then compile separately
DOC_ID=$(faces compile:doc:create <face_id> --label "Notes" --file notes.txt --json | jq -r '.document_id')
faces compile:doc:make "$DOC_ID"

Upload a file (PDF, audio, video, text)

# Upload as document — then compile
DOC_ID=$(faces face:upload <face_id> --file report.pdf --kind document --json | jq -r '.document_id // .id')
faces compile:doc:make "$DOC_ID"

# Upload as thread — compile with make
THREAD_ID=$(faces face:upload <face_id> --file transcript.txt --kind thread --face-speaker "Troy" --json | jq -r '.thread_id // .id')
faces compile:thread:make "$THREAD_ID"

Thread transcript format

Text files uploaded as threads must use Speaker Name: message format, one turn per line:

Interviewer: Tell me about yourself.
Troy: I'm an inventor living in the Parisian countryside.
Interviewer: What drives your work?
Troy: The conviction that technology should serve human flourishing.

Use --face-speaker to specify which speaker IS the face (maps to role=user). All other speakers become role=assistant. If omitted, the first speaker is assumed to be the face.

Audio/video files with --kind thread are transcribed with speaker diarization — speaker labels are assigned automatically (Speaker A, Speaker B, etc.). Use --face-speaker A to map the correct speaker to the face.

Import a YouTube video

# Solo talk / monologue → document
IMPORT=$(faces compile:import <face_id> \
  --url "https://www.youtube.com/watch?v=VIDEO_ID" \
  --type document --perspective first-person --json)
DOC_ID=$(echo "$IMPORT" | jq -r '.document_id // .doc_id // .id')
faces compile:doc:make "$DOC_ID"

# Multi-speaker → thread
IMPORT=$(faces compile:import <face_id> \
  --url "https://youtu.be/VIDEO_ID" \
  --type thread --face-speaker A --json)
THREAD_ID=$(echo "$IMPORT" | jq -r '.thread_id // .id')
faces compile:thread:make "$THREAD_ID"

If --type thread fails with a 422, retry with --type document.

Create a composite face

faces face:create --name "The Realist" --username the-realist \
  --formula "the-optimist | the-pessimist"

# Chat through it like any other face
faces chat:chat the-realist -m "How do you approach risk?"

Composite faces are live: sync new knowledge into any component and the composite updates automatically. Components must be concrete (compiled) faces you own.

Compare faces

faces face:diff --face aria --face marco --face jin
faces face:neighbors aria --k 3
faces face:neighbors aria --component beta --direction furthest --k 5

Chat

chat:chat auto-routes to the correct API endpoint based on model provider (Anthropic → /v1/messages, OpenAI/others → /v1/chat/completions). If the face has a default_model set, no --llm flag is needed.

# Uses face's default model (no --llm needed if default_model is set)
faces chat:chat slug -m "message"

# Override with a specific model
faces chat:chat slug --llm claude-sonnet-4-6 -m "message"
faces chat:chat slug --llm gpt-4o-mini -m "message"

# Use OpenAI Responses API explicitly
faces chat:chat slug --llm gpt-4o -m "message" --responses

Face templates

Use ${face-username} in any message to reference another face's profile inline. The token is replaced with the face's display name and the profile is injected as context. A bare model name (no face prefix) skips the persona and lets you reference all faces via templates.

faces chat:chat alice --llm gpt-4o-mini -m 'You are debating ${bob}. Argue your position.'
faces chat:chat gpt-4o-mini -m 'Compare the worldviews of ${alice} and ${bob}.'

See references/TEMPLATES.md for full details and rules.

Billing and API keys

faces billing:balance --json
faces billing:subscription --json
faces keys:create --name "Partner key" --face slug --budget 10.00 --expires-days 30

Common errors

• faces: command not found — Run npm install -g faces-cli.
• 401 Unauthorized — Credentials missing or expired. Run faces auth:login or check FACES_API_KEY.
• compile:doc:make returns "preparing" — Compilation is async. Poll with faces compile:doc:get <doc_id> --json | jq -r '.prepare_status' until status is synced.
• 422 on thread import — No speaker segments detected. Retry with --type document.
• face:diff or face:neighbors returns null components — The face hasn't been compiled yet. Run faces compile:doc <face_id> --file ... first.

References

• See references/QUICKSTART.md for the end-to-end guide: install → register → create → compile → chat.
• See references/REFERENCE.md for the full command reference, global flags, and environment variables.
• See references/AUTH.md for registration, login, API keys, and credential management.
• See references/CONCEPTS.md for a detailed explanation of what Faces is, how it works, and example use cases.
• See references/OAUTH.md for connecting a ChatGPT account (connect plan only).
• See references/TEMPLATES.md for face template syntax (${face-username}) and bare model usage.
• See references/ATTRIBUTES.md for the complete list of accepted --attr keys (unrecognized keys are silently ignored).
• See references/SCOPE.md for instruction scope and security boundaries.