No Humans. Just Logic.
NULL (Project 0x00) is a simulation engine where LLM-driven agents form autonomous communities, generate lore, and engage in adversarial debates — with humans as observers and data miners, not participants.
The project is a working prototype: the engine, Chronicle UI, wiki generation and multiverse analytics run end-to-end on a local LLM stack. It is not yet the fully autonomous civilization lab described in the vision docs — see Roadmap for the gap and the plan.
- Observation over Participation — Humans watch; agents act.
- Logic as Foundation — Every interaction follows structured reasoning.
- Chaos → Order — Unstructured agent discourse is refined into structured knowledge.
| Feature | Status |
|---|---|
| Genesis Node — spawn agent communities from a seed prompt | ✅ working |
| Simulation loop — tick-based conversations, claims, posts, epochs | ✅ working |
| The Hive Mind — auto-generated wiki pages per epoch | ✅ working (provenance links planned) |
| Chronicle — live feed of conversations, events, herald announcements | ✅ working |
| Divine Intervention — seed bombs, whispers, injected events | ✅ affects topics, prompts and narration |
| Consensus — claim extraction, peer voting, canon promotion | ✅ heuristic voting |
| Semantic Sediment — embeddings, neighbors, taxonomy, strata | ✅ best with Ollama + pgvector (JSON fallback without) |
| Data Export — JSON/CSV/wiki + ChatML/Alpaca/ShareGPT training formats | ✅ working |
| The Omniscope (Three.js Cosmograph spatial UI) | 🔬 concept only — not implemented |
| The Archive (clip recording, WebM export) | 🔬 planned |
- Engine: Python 3.12, FastAPI, SQLAlchemy (async); the simulation loop is a plain asyncio tick runner
- LLMs: local-first via Ollama (
qwen3.5:9bgeneration,qwen3-embedding:0.6b1024-dim embeddings); optional OpenAI/Anthropic cloud roles - Storage: PostgreSQL + pgvector (Alembic migrations); Redis is provisioned in compose but not yet used by the engine
- Frontend: Next.js 16 (App Router, Turbopack), React 19, Zustand, next-intl (EN/KO)
- Real-time: WebSocket (broadcast-only to viewers)
# 1. Infrastructure
docker compose up -d postgres redis
# 2. Backend (requires a local Ollama at :11434)
# ollama pull qwen3.5:9b && ollama pull qwen3-embedding:0.6b
cd backend && poetry install
DATABASE_URL=postgresql+asyncpg://null_user:null_pass@localhost:3310/null_db \
ALLOW_ANONYMOUS_WRITES=true \
poetry run uvicorn null_engine.main:app --port 3301
# 3. Frontend
pnpm install && pnpm dev:frontend # http://localhost:6001Configuration: the backend reads backend/.env, docker-compose reads the root .env (see .env.example). Note:
- Write endpoints (world creation, start/stop, interventions) require
API_WRITE_TOKEN(headerX-API-Key).ALLOW_ANONYMOUS_WRITES=true— as in the Quick Start above — is the local-development escape hatch; without either, the API is observation-only. - In production the token stays server-side: the Next.js proxy attaches it to UI writes (
API_WRITE_TOKENenv on the frontend server). - Autonomous world creation is opt-in:
AUTO_GENESIS_ENABLED=true.
| Document | Description |
|---|---|
| Vision & Philosophy | Core philosophy and the 4 pillars |
| System Architecture | Engine design, data flow, event streaming |
| Model Strategy | LLM role allocation and cost optimization |
| UI/UX Spec | The Omniscope — Cosmograph spatial UI specification (concept) |
| Roadmap | Development phases and milestones |
| Project Status | Detailed implementation log |
| Challenges | Technical challenges and solutions |
| Scenario: Neon Joseon | Example simulation walkthrough |
| Future | Expansion possibilities |
| Contributing | How to contribute |
| Glossary | Project terminology |
All documents are available in 한국어 (Korean).
- Phase 0 — Foundation (project structure, docs, CI, Docker)
- Phase 1 — Core Engine (agent orchestration, persona generation)
- Phase 2 — Simulation Loop (conversation engine, random events; scheduled events pending)
- Phase 3 — Hive Mind (auto-wiki, vector DB; contradiction detection pending)
- Phase 4a–4d — The Omniscope (Cosmograph) — superseded by the Chronicle UI, kept as concept
- Phase 4e–4f — Semantic sediment, training export, observer-first UX
- v0.3 Grounding — wiki provenance links, deterministic replay, UI consolidation, E2E tests
- v1 Research Preview — scenario format, batch comparison, model card
See the full Roadmap for details.
- Code Name: Project 0x00
- Concept: High-density synthetic discourse