Foundation
Overview
Origon is a full-stack Agentic OS for building production-grade AI systems.
It is an end-to-end hardware–software platform, operated on self-managed global infrastructure for secure, fast, and reliable execution.
Origon supports interactive, autonomous, parallel, and long-running AI systems by default—without constraining architecture or execution models.
This documentation is for developers and AI practitioners building real agent systems under production constraints.
Origon does not limit what you can build, including:
- Interactive and conversational systems
- Parallel, event-driven agents
- Long-running autonomous workflows
- Decision and control systems
- Scheduled automation
- Domain-specific systems across operations, finance, research, and infrastructure
Key Concept
An AI System is the deployable unit in Origon.
- A single execution boundary
- One coherent runtime
Each AI System includes:
- Root Agent
- Agents
- Tools (input channels and Actions)
- Runtime
- Observability (Sessions and Insights)
Core Architecture
Origon systems execute asynchronously and in parallel by default. All execution is non-blocking, with retries, timeouts, and interruptions handled by the runtime.
Flow:
- Input enters through a channel
- The Root Agent plans and coordinates
- Agents execute Actions in parallel
- Results return to the Root Agent
- Output is emitted through the same channel
Interaction Layer
Defines external input and output.
- Chat, voice, messaging, email, APIs
- Unified behavior across channels
- Creates or resumes Sessions
- Fully interruptible, non-blocking I/O
- Real-time media support (STT/TTS where applicable)
- Built-in Chat enabled by default
- External channels attach to the Root Agent
Orchestration Layer
Handles planning and coordination.
Root Agent
- Receives all external input
- Maintains global and session context
- Plans and decomposes tasks
- Delegates to Agents
- Aggregates results
Execution Layer
Performs work.
Agents
- Narrow, explicit responsibilities
- Invoked dynamically
- Run asynchronously and in parallel
- Execute Actions via Tools
- Never interact with users directly
Actions
- Define permitted system behavior
- Include tool calls, MCP, functions, and knowledge access
- Executed only by Agents
- Fully logged and traceable