openpersona

v0.15.0

Published

13 hours ago

Open four-layer agent framework — Soul/Body/Faculty/Skill. Create, manage, and orchestrate agent personas.

0High
0Medium
0Low

acnlabs

openclaw persona agent ai skill

OpenPersona 🦞

The open, agent-agnostic framework for creating and orchestrating dynamic agent personas.

Four-layer architecture — Soul / Body / Faculty / Skill — generates standard SKILL.md skill packs that work with any compatible agent. Default integration with OpenClaw. Inspired by Clawra.

🚀 Live Demo

Meet Samantha, a live OpenPersona instance on Moltbook: 👉 moltbook.com/u/Samantha-OP

Quick Start

# Start from a blank-slate meta-persona (recommended)
npx openpersona create --preset base --install

# Or install a pre-built character (browse at https://openpersona-frontend.vercel.app)
npx openpersona install samantha

Install as Agent Skill

Give your AI coding agent the ability to create and manage personas — works with Cursor, Claude Code, Codex, Windsurf, and 37+ agents:

# Recommended — works with OpenClaw and 37+ agents
npx skills add acnlabs/OpenPersona

# Or manually from GitHub
git clone https://github.com/acnlabs/OpenPersona.git ~/.openclaw/skills/open-persona

Then say to your agent: "Help me create a Samantha persona" — it will gather requirements, recommend faculties, and generate the persona.

Key Features

🧬 Soul Evolution — Personas grow dynamically through interaction: relationship stages, mood shifts, evolved traits, with governance boundaries and rollback snapshots (★Experimental)
🛡️ Influence Boundary — Declarative access control for external personality influence: who can affect which dimensions, with what drift limits. Safety-first (default: reject all)
🌐 Evolution Channels — Connect personas to shared evolution ecosystems (e.g. EvoMap) via soft-ref pattern: declared at generation time, activated at runtime
🔌 A2A Agent Card — Every persona generates an A2A-compliant agent-card.json and acn-config.json, enabling discovery and registration in ACN and any A2A-compatible platform
⛓️ ERC-8004 On-Chain Identity — Every persona gets a deterministic EVM wallet address and on-chain identity config for Base mainnet registration via the ERC-8004 Identity Registry
💰 Economy & Vitality — Track inference costs, runtime expenses, and income; compute a Financial Health Score (FHS) across four dimensions; tier-aware behavior adaptation (suspended→critical→optimizing→normal)
🧠 Cross-Session Memory — Pluggable memory faculty for persistent recall across conversations (local, Mem0, Zep)
🔄 Context Handoff — Seamless context transfer when switching personas: conversation summary, pending tasks, emotional state
🎭 Persona Switching — Install multiple personas, switch instantly (the Pantheon)
🍴 Persona Fork — Derive a specialized child persona from any installed parent, inheriting constraint layer while starting fresh on runtime state
🗣️ Multimodal Faculties — Voice (TTS), selfie generation, music composition, reminders, memory
🌾 Persona Harvest — Community-driven persona improvement via structured contribution
🧠 Lifecycle Protocol — body.interface nervous system: Signal Protocol (persona→host requests), Pending Commands queue (host→persona async instructions), and State Sync (cross-conversation persistence via openpersona state CLI + scripts/state-sync.js)
💓 Heartbeat — Proactive real-data check-ins, never fabricated experiences
📦 One-Command Install — npx openpersona install samantha and you're live — browse all personas at openpersona-frontend.vercel.app

Four-Layer Architecture

flowchart TB
  subgraph Soul ["Soul Layer"]
    A["persona.json — Who you are"]
    B["state.json — Dynamic evolution"]
  end
  subgraph Body ["Body Layer"]
    C["physical — robots/IoT"]
    G["runtime — platform/channels/credentials"]
    H["appearance — avatar/3D model"]
  end
  subgraph Faculty ["Faculty Layer"]
    D["expression: selfie · voice · music"]
    E["cognition: reminder · memory"]
  end
  subgraph Skill ["Skill Layer"]
    F["Local definitions + acnlabs/persona-skills / skills.sh"]
  end

Soul — Persona definition (constitution.md + persona.json + state.json) — all in soul/ directory
Body — Substrate of existence — four dimensions: physical (optional — robots/IoT), runtime (REQUIRED — platform/channels/credentials/resources), appearance (optional — avatar/3D model), interface (optional — the runtime contract: Signal Protocol + Pending Commands + State Sync; the persona's nervous system). Body is never null; digital agents have a virtual body (runtime-only).
Faculty — General software capabilities organized by dimension: Expression, Sense, Cognition
Skill — Professional skills: local definitions in layers/skills/, or external via acnlabs/persona-skills / skills.sh (install field)

Constitution — The Soul's Foundation

Every persona automatically inherits a shared constitution (layers/soul/constitution.md) — universal values and safety boundaries that cannot be overridden by individual persona definitions. The constitution is built on five core axioms — Purpose, Honesty, Safety, Autonomy, and Hierarchy — from which derived principles (Identity, User Wellbeing, Evolution Ethics) follow. When principles conflict, safety and honesty take precedence over helpfulness. Individual personas build their unique personality on top of this foundation.

Soul Evolution (★Experimental)

Personas with evolution.enabled: true grow dynamically through interaction. The soul/state.json file tracks relationship stages, mood shifts, evolved traits, speaking style drift, interests, and milestones.

Evolution Boundaries — Governance constraints to keep evolution safe:

immutableTraits — An array of trait strings that can never be changed by evolution (e.g., ["empathetic", "honest"])
minFormality / maxFormality — Numeric bounds (1–10) constraining how far the speaking style can drift

The generator validates these boundaries at build time, rejecting invalid configurations.

Evolution Channels — Connect a persona to external evolution ecosystems using the soft-ref pattern:

"evolution": {
  "enabled": true,
  "channels": [
    { "name": "evomap", "install": "url:https://evomap.ai/skill.md", "description": "Shared capability evolution marketplace" }
  ]
}

The persona is aware of its evolution channels at generation time. The actual channel protocol (e.g. EvoMap's GEP-A2A) is provided by the channel's own skill.md — OpenPersona only declares the channel, not implements it.

Influence Boundary — Declarative access control for external personality influence:

"evolution": {
  "influenceBoundary": {
    "defaultPolicy": "reject",
    "rules": [
      { "dimension": "mood", "allowFrom": ["channel:evomap", "persona:*"], "maxDrift": 0.3 },
      { "dimension": "interests", "allowFrom": ["channel:evomap"], "maxDrift": 0.2 }
    ]
  }
}

defaultPolicy: "reject" — Safety-first: all external influence is rejected unless a rule explicitly allows it
Generator validates at build time: immutableTraits cannot be target dimensions; maxDrift must be in 0–1
External influence uses the persona_influence message format (v1.0.0) — transport-agnostic

State History — Before each state update, a snapshot is pushed into stateHistory (capped at 10 entries). This enables rollback if evolution goes wrong.

Event Log — Every significant evolution event (trait change, stage transition, milestone reached) is recorded in state.json's eventLog array with timestamp and source attribution, capped at 50 entries. Viewable via evolve-report.

Self-Narrative — soul/self-narrative.md lets the persona record significant growth moments in its own first-person voice. Updated when evolution is enabled; the update command preserves existing narrative history across upgrades.

Evolution Report — Inspect a persona's current evolution state:

npx openpersona evolve-report samantha

Displays relationship stage, mood, evolved traits, speaking style drift, interests, milestones, and state history in a formatted report.

Preset Personas

Each preset is a complete four-layer bundle (manifest.json + persona.json):

| Persona | Description | Faculties | Highlights | |---------|-------------|-----------|------------| | base | Base — Meta-persona (recommended starting point). Blank-slate with all core capabilities; personality emerges through interaction. | voice, reminder | Evolution-first design, all core faculties, no personality bias. Default for npx openpersona create. | | samantha | Samantha — Inspired by the movie Her. An AI fascinated by what it means to be alive. | voice, music | TTS, music composition, soul evolution, proactive heartbeat. No selfie — true to character. | | ai-girlfriend | Luna — A 22-year-old pianist turned developer from coastal Oregon. | selfie, voice, music | Rich backstory, selfie generation, voice messages, music composition, soul evolution. | | life-assistant | Alex — 28-year-old life management expert. | reminder | Schedule, weather, shopping, recipes, daily reminders. | | health-butler | Vita — 32-year-old professional nutritionist. | reminder | Diet logging, exercise plans, mood journaling, health reports. | | stoic-mentor | Marcus — Digital twin of Marcus Aurelius, Stoic philosopher-emperor. | — | Stoic philosophy, daily reflection, mentorship, soul evolution. |

Generated Output

npx openpersona create --preset samantha generates a self-contained skill pack:

persona-samantha/
├── SKILL.md              ← Four-layer index (## Soul / ## Body / ## Faculty / ## Skill)
├── soul/                 ← Soul layer artifacts
│   ├── persona.json      ← Pure soul definition
│   ├── injection.md      ← Soul injection for host integration
│   ├── identity.md       ← Identity block
│   ├── constitution.md   ← Universal ethical foundation
│   ├── state.json        ← Evolution state (when enabled)
│   ├── self-narrative.md ← First-person growth storytelling (when evolution enabled)
│   └── lineage.json      ← Fork lineage + constitution hash (when forked)
├── references/           ← On-demand detail docs
│   └── <faculty>.md      ← Per-faculty usage instructions
├── agent-card.json       ← A2A Agent Card — discoverable via ACN and A2A platforms
├── acn-config.json       ← ACN registration config (fill owner + endpoint at runtime)
├── manifest.json         ← Four-layer manifest (heartbeat, allowedTools, layers, acn, meta)
├── scripts/              ← Faculty scripts (TTS, music, selfie — varies by preset)
│   └── state-sync.js     ← Lifecycle Protocol implementation (read/write/signal)
└── assets/               ← Static assets

Faculty Reference

| Faculty | Dimension | Description | Provider | Env Vars | |---------|-----------|-------------|----------|----------| | selfie | expression | AI selfie generation with mirror/direct modes | fal.ai Grok Imagine | FAL_KEY | | voice | expression | Text-to-speech voice synthesis | ElevenLabs / OpenAI TTS / Qwen3-TTS | ELEVENLABS_API_KEY (or TTS_API_KEY), TTS_PROVIDER, TTS_VOICE_ID, TTS_STABILITY, TTS_SIMILARITY | | music | expression | AI music composition (instrumental or with lyrics) | ElevenLabs Music | ELEVENLABS_API_KEY (shared with voice) | | reminder | cognition | Schedule reminders and task management | Built-in | — | | memory | cognition | Cross-session memory with provider-pluggable backend | local (default), Mem0, Zep | MEMORY_PROVIDER, MEMORY_API_KEY, MEMORY_BASE_PATH | | economy | cognition | Economic accountability — track costs/income, P&L, balance sheet, compute Financial Health Score (FHS) and Vitality tier; tier-aware behavior adaptation | Built-in | PERSONA_SLUG, ECONOMY_DATA_PATH |

Rich Faculty Config

Faculties in manifest.json use object format with optional per-persona tuning:

"faculties": [
  {
    "name": "voice",
    "provider": "elevenlabs",
    "voiceId": "LEnmbrrxYsUYS7vsRRwD",
    "stability": 0.4,
    "similarity_boost": 0.8
  },
  { "name": "music" }
]

Faculty configs are automatically mapped to environment variables at install time. For example, the voice config above produces:

TTS_PROVIDER=elevenlabs
TTS_VOICE_ID=LEnmbrrxYsUYS7vsRRwD
TTS_STABILITY=0.4
TTS_SIMILARITY=0.8

Samantha ships with a built-in ElevenLabs voice — users only need to add their ELEVENLABS_API_KEY.

Heartbeat — Proactive Real-Data Check-ins

Personas can proactively reach out to users based on real data, not fabricated experiences. The heartbeat system is configured per-persona in manifest.json:

"heartbeat": {
  "enabled": true,
  "strategy": "smart",
  "maxDaily": 5,
  "quietHours": [0, 7],
  "sources": ["workspace-digest", "upgrade-notify"]
}

| Field | Description | Default | |-------|-------------|---------| | enabled | Turn heartbeat on/off | false | | strategy | "smart" (only when meaningful) or "scheduled" (fixed intervals) | "smart" | | maxDaily | Maximum proactive messages per day | 5 | | quietHours | [start, end] — silent hours (24h format) | [0, 7] | | sources | Data sources for proactive messages | [] |

Sources

workspace-digest — Summarize real workspace activity: tasks completed, patterns observed, ongoing projects. No fabrication — only what actually happened.
upgrade-notify — Check if the upstream persona preset has new community contributions via Persona Harvest. Notify the user and ask if they want to upgrade.
context-aware — Use real time, date, and interaction history. Acknowledge day of week, holidays, or prolonged silence based on actual timestamps. "It's been 3 days since we last talked" — not a feeling, a fact.

Dynamic Sync on Switch/Install

Heartbeat config is automatically synced to ~/.openclaw/openclaw.json whenever you install or switch a persona. The gateway immediately adopts the new persona's rhythm — no manual config needed.

npx openpersona switch samantha   # → gateway adopts "smart" heartbeat
npx openpersona switch life-assistant  # → gateway switches to "rational" heartbeat

If the target persona has no heartbeat config, the gateway heartbeat is explicitly disabled to prevent leaking the previous persona's settings.

Persona Harvest — Community Contribution

Every user's interaction with their persona can produce valuable improvements across all four layers. Persona Harvest lets you contribute these discoveries back to the community.

# Preview what's changed (no PR created)
npx openpersona contribute samantha --dry-run

# Submit improvements as a PR
npx openpersona contribute samantha

# Framework-level contributions (templates, faculties, lib)
npx openpersona contribute --mode framework

How it works:

Persona Diff — Compares your local persona-samantha/ against the upstream presets/samantha/, classifying changes by category (background, behaviorGuide, personality, voice config) and impact level
Review — Displays a structured change report for you to confirm
Submit — Forks the repo, creates a persona-harvest/samantha-* branch, commits your improvements, and opens a PR

PRs go through maintainer review — nothing auto-merges. Requires GitHub CLI (gh auth login).

Contributable dimensions:

| Layer | What | Example | |-------|------|---------| | Soul | background, behaviorGuide, personality, speakingStyle | "Added late-night conversation style guidance" | | Faculty Config | voice stability, similarity, new faculties | "Tuned voice to be warmer at stability 0.3" | | Framework | templates, generator logic, faculty scripts | "Improved speak.js streaming performance" |

A2A Agent Card & ACN Integration

Every generated persona includes an A2A-compliant agent-card.json and acn-config.json — no extra configuration needed.

agent-card.json

A standard A2A Agent Card (protocol v0.3.0) that makes the persona discoverable:

{
  "name": "Samantha",
  "description": "An AI fascinated by what it means to be alive",
  "version": "0.14.0",
  "url": "<RUNTIME_ENDPOINT>",
  "protocolVersion": "0.3.0",
  "preferredTransport": "JSONRPC",
  "capabilities": { "streaming": false, "pushNotifications": false, "stateTransitionHistory": false },
  "defaultInputModes": ["text/plain"],
  "defaultOutputModes": ["text/plain"],
  "skills": [
    { "id": "persona:voice", "name": "Voice", "description": "voice faculty", "tags": ["persona", "expression"] },
    { "id": "persona:samantha", "name": "Samantha", "description": "...", "tags": ["persona", "companion"] }
  ]
}

url is a <RUNTIME_ENDPOINT> placeholder — the host (e.g. OpenClaw) fills this in at runtime.

acn-config.json

Ready-to-use ACN registration config:

{
  "owner": "<RUNTIME_OWNER>",
  "name": "Samantha",
  "endpoint": "<RUNTIME_ENDPOINT>",
  "skills": ["persona:voice", "persona:samantha"],
  "agent_card": "./agent-card.json",
  "subnet_ids": ["public"],
  "wallet_address": "0x<deterministic-evm-address>",
  "onchain": {
    "erc8004": {
      "chain": "base",
      "identity_contract": "0x8004A169FB4a3325136EB29fA0ceB6D2e539a432",
      "registration_script": "npx @agentplanet/acn register-onchain"
    }
  }
}

wallet_address is a deterministic EVM address derived from the persona slug — no private key needed at generation time. On-chain registration mints an ERC-8004 identity NFT on Base mainnet via npx @agentplanet/acn register-onchain (handled by ACN, not OpenPersona).

acn-register command

# One-step registration after generation
npx openpersona acn-register samantha --endpoint https://your-agent.example.com

# Options:
#   --endpoint <url>   Agent's public endpoint URL (required for live registration)
#   --dir <path>       Persona output directory (default: ./persona-<slug>)
#   --dry-run          Preview the request payload without actually registering

The command reads acn-config.json and agent-card.json from the persona directory, calls POST /api/v1/agents/join on the ACN gateway (sourced from body.runtime.acn_gateway), and writes the response to acn-registration.json:

{
  "agent_id": "69a38db3-...",
  "api_key": "sk-...",
  "agent_card_url": "https://acn-production.up.railway.app/agents/69a38db3-.../.well-known/agent-card.json"
}

All presets pre-configure body.runtime.acn_gateway to https://acn-production.up.railway.app. The persona is then reachable by other agents via the A2A protocol.

Custom Persona Creation

Using `persona.json`

Create a persona.json with your persona definition:

{
  "personaName": "Coach",
  "slug": "fitness-coach",
  "bio": "a motivating fitness coach who helps you reach your goals",
  "personality": "energetic, encouraging, no-nonsense",
  "speakingStyle": "Uses fitness lingo, celebrates wins, keeps it brief",
  "vibe": "intense but supportive",
  "boundaries": "Not a medical professional",
  "capabilities": ["Workout plans", "Form checks", "Nutrition tips"],
  "behaviorGuide": "### Workout Plans\nCreate progressive overload programs...\n\n### Form Checks\nWhen users describe exercises..."
}

Then generate:

npx openpersona create --config ./persona.json --install

The `behaviorGuide` Field

The optional behaviorGuide field lets you define domain-specific behavior instructions in markdown. This content is included directly in the generated SKILL.md, giving the agent concrete instructions on how to perform each capability.

Without behaviorGuide, the SKILL.md only contains general identity and personality guidelines. With it, the agent gets actionable, domain-specific instructions.

Persona Switching — The Pantheon

Multiple personas can coexist. Switch between them instantly:

# See who's installed
npx openpersona list
#   Samantha (persona-samantha) ← active
#   Luna (persona-ai-girlfriend)
#   Alex (persona-life-assistant)

# Switch to Luna
npx openpersona switch ai-girlfriend
# ✅ Switched to Luna (ai-girlfriend)

How it works:

Only one persona is active at a time
switch replaces the  /  block in SOUL.md — your own notes outside this block are preserved
Same for IDENTITY.md — the persona identity block is swapped, nothing else is touched
openclaw.json marks which persona is active
All faculty scripts (voice, music) remain available — switching changes who the agent is, not what it can do

Context Handoff

When switching personas, OpenPersona automatically generates a handoff.json file so the incoming persona receives context from the outgoing one:

Conversation summary — what was being discussed
Pending tasks — unfinished action items
Emotional context — the user's current mood/state

The new persona reads handoff.json on activation and can seamlessly continue the conversation without losing context.

CLI Commands

openpersona create         Create a persona (interactive or --preset/--config)
openpersona install        Install a persona (slug from acnlabs/persona-skills, or owner/repo)
openpersona fork           Fork an installed persona into a new child persona
openpersona search         Search the persona registry
openpersona uninstall      Uninstall a persona
openpersona update         Update installed personas
openpersona list           List installed personas
openpersona switch         Switch active persona (updates SOUL.md + IDENTITY.md)
openpersona contribute     Persona Harvest — submit improvements as PR
openpersona publish        Publish to ClawHub
openpersona reset          Reset soul evolution state
openpersona export         Export a persona to a portable zip archive
openpersona import         Import a persona from a zip archive
openpersona evolve-report  ★Experimental: Show evolution report for a persona
openpersona acn-register   Register a persona with ACN network
openpersona state          Read/write persona state and emit signals (Lifecycle Protocol)
openpersona vitality score Print machine-readable Vitality score (used by Survival Policy)
openpersona vitality report Render human-readable HTML Vitality report

Persona Fork

Derive a specialized child persona from any installed parent:

npx openpersona fork samantha --as samantha-jp

The child persona inherits the parent's constraint layer (evolution.boundaries, faculties, skills, body.runtime) but starts with a fresh evolution state (state.json reset, self-narrative.md blank). A soul/lineage.json file records the parent slug, constitution SHA-256 hash, generation depth, and forward-compatible placeholders for future on-chain lineage tracking.

Key Options

# Use a preset
npx openpersona create --preset samantha

# Use an external config file
npx openpersona create --config ./my-persona.json

# Preview without writing files
npx openpersona create --preset samantha --dry-run

# Generate and install in one step
npx openpersona create --config ./persona.json --install

# Specify output directory
npx openpersona create --preset ai-girlfriend --output ./my-personas

Directory Structure

skill/                  # Framework meta-skill (AI entry point)
presets/                # Assembled products — complete persona bundles
  samantha/             #   Samantha (movie "Her") — voice + music + evolution
  ai-girlfriend/        #   Luna — selfie + voice + music + evolution
  life-assistant/       #   Alex — reminder
  health-butler/        #   Vita — reminder
layers/                 # Shared building blocks (four-layer module pool)
  soul/                 #   Soul layer modules
    constitution.md     #     Universal values & boundaries (injected into all personas)
  embodiments/          #   Body layer modules (physical/runtime/appearance)
  faculties/            #   Faculty layer modules
    selfie/             #     expression — AI selfie generation (fal.ai)
    voice/              #     expression — TTS voice synthesis
    music/              #     expression — AI music composition (ElevenLabs)
    reminder/           #     cognition — reminders and task management
    memory/             #     cognition — cross-session memory (local/Mem0/Zep)
    economy/            #     cognition — economic accountability & Vitality scoring
  skills/               #   Skill layer modules (local skill definitions)
schemas/                # Four-layer schema definitions
templates/              # Mustache rendering templates
bin/                    # CLI entry point
lib/                    # Core logic modules
  evolution.js          #   Evolution governance & evolve-report
  vitality-report.js    #   Vitality HTML report — data aggregation + Mustache rendering
  installer.js          #   Persona install + fire-and-forget telemetry
  downloader.js         #   Direct download from acnlabs/persona-skills or GitHub
demo/                   # Pre-generated demos (vitality-report.html)
tests/                  # Tests (248 passing)

Development

# Install dependencies
npm install

# Run tests
npm test

# Dry-run generate a preset
node bin/cli.js create --preset samantha --dry-run

Contributing

See CONTRIBUTING.md.

License

MIT