Introduction
Contenox is a self-hostable runtime for building observable, deterministic AI workflows as explicit state machines.
Instead of wiring prompts together with ad-hoc Python glue, you define your AI behaviour as a task chain — a JSON graph of typed tasks, transitions, and tool calls. Every step is inspectable, replayable, and testable.
Three editions
| Edition | Use case | Entry point |
|---|---|---|
| vibe CLI | Local AI agent on your machine | vibe run |
| Runtime API | Self-hosted REST backend for apps | Docker / go run |
| Enterprise (EE) | Multi-tenant, dashboard, RBAC | enterprise/ |
vibe is the flagship. This documentation focuses on it first. The runtime API and EE share the same chain / hook / task engine underneath.
How it works
User input
│
▼
┌─────────────────────┐
│ Task Chain (JSON) │ ← you define this
│ task → task → … │
└─────────────────────┘
│
▼
Model (Ollama / OpenAI / vLLM / Gemini)
│
├─ tool call? → Hook (local shell, remote API)
│ │
└─ text reply ←──────┘Each task has a handler (what it does), an optional LLM config (which model, which hooks), and a transition (where to go next). The chain engine drives the loop — the model doesn't.
Next steps
- Quickstart — install vibe and run your first chain in 5 minutes
- Core Concepts — chains, tasks, hooks, transitions explained
- Chains reference — build your own chains from scratch
