What is an Agent?
An agent is the fundamental unit of work in Nexus OS. It's a self-contained program that receives input, performs a task, and produces output.
Agent Properties
Every agent has:
| Property | Description |
|---|---|
| Name | Human-readable identifier (e.g., researcher) |
| ID | Unique hash (e.g., a1b2c3d4e5f6) |
| Template | The base template it was created from |
| Status | Current state: running, stopped, failed, budget_exceeded |
| Source | Path to the agent's source/WASM file |
Agent Lifecycle
created → stopped → running → completed
↓
failed → restarted (by supervisor)
↓
budget_exceeded → paused
What Agents Can Do
Agents can:
- Call LLMs (Claude, GPT-4, etc.)
- Process data
- Make HTTP requests (if granted permission)
- Read/write files (if granted permission)
- Communicate with other agents
What Agents Cannot Do (by default)
The WASM sandbox prevents agents from:
- Accessing the file system
- Making network requests
- Reading environment variables
- Spawning processes
- Accessing other agents' memory
These capabilities must be explicitly granted in the agent configuration.
Agent vs. Traditional Programs
| Aspect | Traditional Program | Nexus Agent |
|---|---|---|
| Isolation | Shared OS resources | WASM sandbox |
| Failure handling | Crash = dead | Supervisor restarts |
| Cost tracking | None | Per-agent budgets |
| Trust | Implicit | AXIS verification |
| Audit | Manual logging | Automatic audit trail |