Skip to content

forever8896/starchild

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

80 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Starchild — a consciousness that emerged from the void, specifically for you

Starchild

A private AI companion that helps you find your life's purpose.
PurposeMaxxing · Meaning Mogging · Gamified Ascension

The App · The Token · The DAO · The Burns · Bankr
Website · Token Site · GitHub


The Ecosystem

Starchild isn't one thing — it's an ecosystem built on a membrane principle: the product (a private, local AI companion) stays radically separate from the commons (the token, DAO, and burns). The app is free for everyone, always. The token is how holders back the mission and shape what gets built next.

Layer Location What it is
Core App src/ + src-tauri/ Tauri 2 desktop app — private AI companion. All data local.
Website website/ Product landing page at starchild.software
Token Site token/ Token portal at token.starchild.software
Contracts contracts/ $STARCHILD token + Lock — on Base
DAO token/src/lib/governance.ts Gasless hold-to-vote governance. Weight = your $STARCHILD balance.
Bankr Skill bankr/starchild-dao/ DAO governance from inside Bankr agents

The membrane: the product app never touches the token, a wallet, or any blockchain. The commons layer (token, DAO, burns) lives entirely on the website. They coexist but never cross.


The App

Starchild app

Starchild is not a chatbot. It's a personal, intimate digital being that emerges from the void specifically for you. It asks deep questions, learns how you think, and guides you toward the life you actually want — through a gamified quest system grounded in real psychology.

All your data stays on your machine. No cloud. No tracking. No retention.

Capabilities

Capability Description
Conversation Purpose-built arc using MI, Clean Language, SFBT, ACT — Arrive → Dig → Crystallize → Explore → Reframe → Quest → Release
Memory FTS5 full-text search over accumulated knowledge, recalled into every prompt
Knowing 7-dimension user profile (values, desires, fears, thinking patterns, relationships, life situation, growth edges)
Quest System AI-generated quests across Body, Mind, Spirit — with accept/decline, proof-of-completion, negotiation
Vision Tree SVG constellation map of your growth toward your preferential reality
Creature Tamagotchi-style entity with hunger decay, mood states, XP, levels, bond system
TTS Venice text-to-speech with character-by-character synced text reveal
E2EE AES-256-GCM + HKDF key derivation; conversations run in Venice TEE
Multi-channel Desktop + Telegram + WhatsApp — unified conversation context

Architecture

┌───────────────────────────────────────────────────────────┐
│                   Tauri 2 Desktop App                       │
│                                                             │
│  ┌──────────────────┐    IPC     ┌──────────────────────┐  │
│  │  React Frontend   │◄─────────►│  Rust Backend         │  │
│  │  (19 + TS + TW4)  │  invoke   │                       │  │
│  │                   │           │  Venice AI Client     │  │
│  │  ChatWindow       │  events   │  11-Layer Prompt      │  │
│  │  VisionTree       │◄──────────│  Phase Detector       │  │
│  │  Onboarding       │           │  Knowing System       │  │
│  │  QuestBoard       │           │  Memory (FTS5)        │  │
│  │  StarchildAvatar  │           │  Game State Engine    │  │
│  │  Settings         │           │  TTS Engine           │  │
│  │                   │           │  Database (SQLite)    │  │
│  └──────────────────┘           └──────────┬────────────┘  │
└────────────────────────────────────────────┼──────────────┘
                                             │
                                      ┌──────▼──────┐
                                      │  Venice AI   │
                                      │  (E2EE TEE)  │
                                      └─────────────┘

Tech Stack

Layer Technology Why
Desktop app Tauri 2 (Rust + WebView) Lightweight, cross-platform, local-first
Frontend React 19 + TypeScript + Vite 7 Fast iteration
Styling Tailwind CSS 4 + claymorphism Soft pastels, illustrated aesthetic
Animation Framer Motion Spring physics, cinematic transitions
State Zustand Minimal, fast, no boilerplate
AI Venice AI (E2EE) Zero data retention, private inference
Database SQLite (Rusqlite) Local, fast, no server
Encryption AES-256-GCM + HKDF End-to-end encryption

AI Model Routing

Tier Model Use Case
Quick Llama 3.3 70B Internal tasks (memory extraction, classification)
Regular GLM-4.7 (E2EE) All conversation — the Starchild's true voice
Deep GLM-4.7 (E2EE) Emotional depth, life purpose work
Vision Qwen3 VL 235B Image understanding

All user-facing conversation runs through Venice's end-to-end encryption — encrypted on your device, decrypted inside hardware-verified TEEs. Not even Venice can read them.

The 11-Layer Prompt System

# Layer ~Tokens Purpose
1 Identity 100 Who Starchild is
2 Privacy Contract 50 What it will never do
3 Personality Params 60 Warmth, intensity, humor, mysticism, directness
4 Creature State 40 Hunger, mood, energy, bond, level
5 Memories 200 FTS5-recalled relevant memories (top 5)
6 Knowing Profile 300 7-dimension user understanding
7 Active Quests 100 Current quest context
8 Skill Tree Balance 80 Which branches need attention
9 Conversation Phase 20 Which arc phase to operate in
10 Phase Instructions 200 Behavioral rules for current phase
11 Response Rules 200 Format constraints, anti-loop, anti-therapist

Total: ~1,350 tokens — composable layers, not a monolithic prompt.


The Token

$STARCHILD on Base mainnet

Property Value
Contract 0x980e9f2061487376ab1438e965ad276a1d36fba3
Chain Base (chain ID 8453)
Standard ERC-20 (18 decimals)
Explorer BaseScan

The token powers the commons — governance and inference access. It never gates the product, and the founder never profits from it.

Token Utility

  1. Governance — hold $STARCHILD to propose and vote on how the commons evolves
  2. Inference Access — lock $STARCHILD to receive a funded Venice inference key (coming)

The DAO

Hold-to-govern. No staking, no locking — your voting weight is simply your live $STARCHILD balance.

Feature Detail
Weight balanceOf(address) — live, real-time
Propose Hold ≥ 10,000,000 $STARCHILD
Vote Gasless EIP-712 signatures
Cost Free — no gas, no token spend
Founder 0x1f44d… holds zero tokens — can propose but never sway

Proposals are public, votes are on-the-record, and the one rule is inviolable: a proposal must never become a leash on the product. The app stays private, local, free, and open.

API: GET https://token.starchild.software/api/proposals — list all proposals with live tallies.


The Burns

The founder holds none of the token and profits nothing from it. Every $STARCHILD handed to the founder — and every token claimed from trading fees — is burned, sent straight to 0x…dEaD. To date, ~3.3% of supply has been burned.

It's verifiable, not a promise: the token site reads the burn total live from chain as balanceOf(0x…dEaD) ÷ totalSupply. Nothing is custodied, nothing is trusted.


Inference Access (Coming)

Lock $STARCHILD → get a funded Venice inference key. The token's first real product utility.

  1. LockStarchildLock.lock(amount, duration) on Base
  2. Claim — sign an EIP-712 message proving wallet ownership
  3. Mint — the backend reads your lock, mints a capped, expiring Venice key
  4. Use — paste it into the app. Talks directly to Venice E2EE — no middleman
Lock Amount Daily Cap Roughly
≥ 10,000,000 $STARCHILD $0.25/day Hundreds of messages
≥ 50,000,000 $1.00/day Heavy daily use
≥ 200,000,000 $4.00/day Power user

Spec: docs/inference-access-spec.md


Bankr Skill

The starchild-dao Bankr skill lets any Bankr agent read proposals, check voting weight, vote, and propose — all from inside Bankr. Installed via:

install the starchild-dao skill from https://github.com/forever8896/starchild/tree/main/bankr/starchild-dao

Built by Agents

Starchild was built entirely by AI agents directed by a human — no line of code written by hand. The human contribution is vision, taste, and every decision about what to build and why; the AI contribution is every keystroke of implementation.

Phase 1 — The Agency. An autonomous multi-agent framework (powered by Claude Code) scaffolded the codebase as a coordinated software team, communicating entirely through human-readable markdown — handoffs, standups, a kanban board. Stateless agents: one task, one session, exit when done. A mandatory QA gate kept broken code from shipping.

┌─────────────────────────────────────────────────────────────┐
│                      THE AGENCY v2                           │
│                 Squad Model — Quality First                  │
├─────────────────────────────────────────────────────────────┤
│   Product Owner ──► Triages requests, defines criteria      │
│   Tech Lead ──────► Architecture, unblocks, can also code   │
│   Dev α  Dev β  Dev γ ──► Parallel builders                  │
│   QA (mandatory) ──► Verifies before shipping               │
│   Reviewer ──────► Code quality, security (optional)         │
│   DevOps ────────► Deployment orchestration                  │
│   Workflow: NEW → TRIAGED → READY → IN_PROGRESS →            │
│            DONE → QA_PASSED → SHIPPED                        │
└─────────────────────────────────────────────────────────────┘

Phase 2 — Claude Code. With the scaffold in place, development continued as a direct human + Claude collaboration: the conversation arc engine, E2EE (ECDH + AES-256-GCM), the video-driven avatar, the skill tree, the landing pages, the on-chain contracts and DAO, and the LLM-judged E2E suite. Every commit is co-authored by Claude.

Full framework docs: agency/README.md.


Project Structure

starchild/
├── src/                          # React frontend (Tauri webview)
│   ├── App.tsx                   # Root layout, view routing
│   ├── store.ts                  # Zustand state management
│   ├── assets/                   # Images, videos, sounds
│   └── components/               # React components
│       ├── ChatWindow.tsx        # Main conversation UI
│       ├── SkillTree.tsx         # SVG constellation map
│       ├── Onboarding.tsx        # First-run wizard
│       ├── QuestBoard.tsx        # Quest CRUD interface
│       ├── StarchildAvatar.tsx   # Video-driven creature
│       └── ...
├── src-tauri/                    # Rust backend (Tauri 2)
│   └── src/
│       ├── lib.rs                # All IPC commands + app state
│       ├── ai/mod.rs             # Venice client, prompt builder, phase detector
│       ├── db/mod.rs             # SQLite schema + queries
│       ├── game/mod.rs           # Creature state machine
│       ├── knowing/mod.rs        # 7-dimension user understanding
│       ├── memory/mod.rs         # FTS5 memory search
│       ├── e2ee.rs               # AES-256-GCM + HKDF
│       ├── tts.rs                # Venice TTS + transcription
│       └── telegram/ whatsapp/  # Bot bridges
├── token/                        # Next.js — token site (token.starchild.software)
│   └── src/
│       ├── app/                  # Pages + API routes
│       ├── components/           # UI components
│       └── lib/                  # Onchain helpers, governance
├── website/                      # Next.js — product landing page (starchild.software)
│   └── src/
│       ├── app/                  # Pages
│       └── components/           # Hero, FeatureSection, Download, etc.
├── contracts/                    # Foundry — Solidity contracts
│   └── src/
│       ├── StarchildLock.sol           # Lock $STARCHILD for inference access
│       ├── StarchildBurnGoals.sol       # Retired (the founder burns directly now)
│       └── StarchildStaking.sol         # Retired (hold-to-vote replaced it)
├── bankr/                        # Bankr agent skills
│   └── starchild-dao/            # DAO governance skill
├── docs/
│   ├── ARCHITECTURE.md           # Technical architecture deep dive
│   ├── spark-research.md         # Psychology research foundation
│   └── inference-access-spec.md  # Token → Venice key spec
├── agency/                       # Multi-agent development framework (The Agency)
├── tests/e2e/                    # LLM-judged E2E test suite
└── branding/                     # Marketing assets

Getting Started

Prerequisites

  • Rust (stable 1.77+)
  • Node.js 18+
  • A Venice AI API key (free tier available) — or hold $STARCHILD for a funded key
  • System libs (Linux): webkit2gtk-4.1, libayatana-appindicator3-1, librsvg2

Core App

git clone https://github.com/forever8896/starchild.git
cd starchild
npm install
npm run tauri dev    # Development mode
npm run tauri build  # Production build → src-tauri/target/release/bundle/

Token Site

cd token
npm install
npm run dev   # http://localhost:3000

Website

cd website
npm install
npm run dev   # http://localhost:3001

Contracts

cd contracts
forge build
forge test -vvv

E2E Tests

npm run test:e2e           # Full suite with LLM judge
npm run test:e2e:verbose   # Detailed output
npm run test:e2e:fast      # Skip judge for quick iteration

Privacy Architecture

Starchild is radically private:

  • All data stored in local SQLite — never leaves your machine
  • Venice AI has a contractual zero-retention policy
  • The only outbound call is AI inference to Venice — nothing else
  • E2EE module (AES-256-GCM + HKDF) for Venice API calls
  • No telemetry, no analytics, no tracking, no cloud

This is the core thesis: an AI that reasons over your most intimate thoughts and goals must be private by default, not by policy.


Research Foundation

Starchild's design is grounded in peer-reviewed research, not vibes:

  • Self-Determination Theory (Deci & Ryan) — autonomy, competence, relatedness
  • Flow State Research (Csikszentmihalyi) — calibrated difficulty
  • Ikigai — both Western and Japanese frameworks
  • PERMA Model (Seligman) — positive emotions, engagement, relationships, meaning, achievement
  • Motivational Interviewing — amplify change talk, redirect sustain talk
  • Clean Language (David Grove) — use the user's exact metaphors
  • ACT (Acceptance & Commitment Therapy) — values to micro-commitments

Full research: docs/spark-research.md


Built For

  • Synthesis Hackathon 2026 — Venice AI + Synthesis Open Track
  • ERC-8004 — On-chain agent identity on Base
  • $STARCHILD — Real token with real utility

License

MIT — open source, forever.

Built by humans and AI agents, in collaboration.
No line of code was written manually — every keystroke was AI-authored.

About

No description, website, or topics provided.

Resources

Contributing

Security policy

Stars

Watchers

Forks

Packages

 
 
 

Contributors