# Security and trust

Clark Agent is built for practical automation with a strong emphasis on isolation and traceability.

## Sandbox isolation

- Each task runs in its own disposable sandbox.
- Workspaces can be observed and inspected while tasks run.
- Sandboxes help separate one task's execution from another.

## Typed contracts

- Clark Agent's tool interfaces are typed and schema-driven.
- The product emphasizes explicit tool contracts instead of informal parsing.

## Open-source core

- Clark Agent has an open-source core, which makes the architecture easier to inspect and reason about.

## Trust model

Clark Agent is intended for users who want a more inspectable, execution-oriented AI system rather than a black-box assistant with hidden behavior.