“An operator running four AI runtimes cannot hold the topology in memory.”
STARGAZER
ARCHITECTURE / ENVIRONMENT · INTELLIGENT ENVIRONMENTS
[ GOVERNING LAW ]
TELOS_02: POLY-MODEL ORCHESTRATION
Freedom inside the laboratory. STARGAZER isolates cognitive work from execution, permitting rapid model substitution (JANUS) without architectural drift. A proving ground for intelligence.
Instrument
STARGAZER is an intelligent environment and local-first command hull designed to make complex agentic and local model runtimes spatially governable. It is not a single web application, but a panoramic cockpit that mounts autonomous agent swarms, tool chambers, and live developer CLI runtimes into one unified topology.
Spatial Command
- Unified panoramic hull topology
- 4+1 spatial column layout providing persistent co-presence
- Bounded keyboard/pointer navigation
- Integrated resource telemetry (CPU/RAM per node)
Model Registry
- Local-first inference via Ollama
- Role-based tiering (Imagine, Shape, Refine)
- Support stack (Utility, Embedding)
- Instant hot-swapping via <ERBERUS chamber
Architecture
STARGAZER orchestrates a multi-runtime composition. A Python API server drives the static hull, while a Node PTY server persists terminal sessions over WebSocket. Docker Compose manages the underlying agent proxy networks.
DOCKER ORCHESTRATION
Proxy orchestrator · Orchestra network · Agent Zero containers (Zano/Nexus/Nyx)
DURABLE SESSIONS
TESSERACT node terminal server · WebSocket registry · Ring-buffer replay · xterm.js
LOCAL INFERENCE
Ollama service (Port 11434) · Multi-model deployment · <ERBERUS live registry reconciliation
Engineering
What was difficult. What decisions mattered.
Deterministic Local Boot Composition
STARGAZER is a real operational machine with deep local daemon and container dependencies, not a static mock. The boot sequence systematically stops stale processes, drives Docker Compose to spin up proxy networks, launches background Node and Python servers, and prepares the browser viewport.
Multi-Runtime Convergence
Unifying autonomous containerized agents, structured local-model convergence pipelines, and interactive CLI coding agents (Claude Code CLI / WSL Hermes) into one seamless command desk without tab fatigue.
Functional Model Decomposition
Relying on a single massive model for all tasks is inefficient. STARGAZER implements multi-model tiering via local open-weights (DeepSeek, Qwen, Gemma, Nomic), decomposing reasoning, drafting, and embedding into dedicated pathways managed by the <ERBERUS chamber.
Evidence
What exists
- ·Localhost-first command hull
- ·Four-panel mounted environment
- ·Durable PTY terminal sessions (TESSERACT)
- ·<ERBERUS AI chamber with live model registry
- ·Docker proxy infrastructure
- ·Keyboard navigation system
What does not exist
- ·Public availability
- ·Git repository
- ·File intake broker
- ·Input lease protocol
Stack
Current State
Maturity
VERIFIED LOCAL SYSTEM
Current State
Intentional internal operational environment
Last Verified
2026-07
Repository
Private (Zentraux Group)
Demo
Internal system
Source code is maintained in a private Zentraux Group repository. Architecture, implementation evidence, demonstrations, and selected technical artifacts are presented here.