> ## Documentation Index
> Fetch the complete documentation index at: https://docs.jhansi.io/llms.txt
> Use this file to discover all available pages before exploring further.

# System Architecture

> jhansi.io separates agent interaction, runtime control and workload execution.

<img src="https://mintcdn.com/jhansiio/fgRpH4IQvQdQ3XHj/images/architecture.svg?fit=max&auto=format&n=fgRpH4IQvQdQ3XHj&q=85&s=ebcd6e9bae7fe343f41ffa9e21ae84c9" alt="jhansi.io architecture" width="680" height="480" data-path="images/architecture.svg" />

## Layers

**Agent or application** — chooses the task and sends an execution request.

↓ SDK / MCP

**jhansi.io Runtime** — owns provisioning, execution, results and cleanup.

↓ controlled runtime interface

**Sandbox** — runs the workload inside the configured isolation boundary.

↓ authorised network or credential path

**External systems** — APIs, databases, repositories and internal services.

## Two boundaries

**Control boundary** — the agent can request supported actions, but it does not provision infrastructure directly.

**Execution boundary** — the workload runs inside a sandbox rather than inside the agent client or developer host.

## Future credential boundary

jhansi.io's long-term direction adds a third boundary: access to external systems should be mapped by the runtime without placing reusable secrets in the agent conversation.