The Closed Loop

The product is the loop, not any single project:

discover → install → authenticate → deploy a model → chat and run agents → control industrial equipment → verify and support

Each segment is owned by a specific set of projects. If a segment is broken, the platform is not finished.

Segment by segment

#	Segment	What happens	Projects
1	Discover	A potential user finds the ecosystem, understands its philosophy, and picks an entry point	docs.celestia.world (this site), celestia-island.github.io, e.celestia.world
2	Install	The user gets a running system: admin panel, desktop/web shell, supervised services	arona (cloud API admin panel), shittim-chest (chat desktop/webUI), malkuth (service supervision)
3	Authenticate	Zero-trust identity: registration (invite-gated), login with rate limiting, API keys, RBAC	kirino (auth primitives and RBAC engine)
4	Deploy a model	Pick a model runtime, deploy it to a node, bind it to a chat backend, meter usage	arona (panel + backends), entelecheia (scepter runtime), plana (metering and pricing)
5	Chat & agents	Talk to models, run multi-agent collaboration, persist conversations, manage memory	shittim-chest (UI and chat), entelecheia (agent orchestration), noa (AI-native version control)
6	Industrial control	Remote operations and protocol brokering: Modbus, S7comm, OPC UA; telemetry and write gates	evernight (protocol broker), aoba (Modbus and data-source CLI)
7	Verify & support	Integration tests on real hardware, supervision and self-healing, usage records, feedback channels	celestia-integration, malkuth, plana (usage records)

Every step is testable. Each segment has a defined acceptance test in [celestia-integration](https://github.com/celestia-island/celestia-integration); a release is not green until the whole loop passes on real nodes.
Every step is observable. Supervision, health endpoints, and usage records make each segment's state visible rather than assumed.
No silent degradation. When a segment degrades (for example memory offline or a backend unreachable), the API response and the UI say so explicitly. Failures are loud by design.
Safety is not a segment. Write gates, policy validation, and human confirmation are woven through segments 5 and 6, not bolted on at the end. See [Safety Principles](./safety.md).