Container Use

Each agent runs in a fresh container with its own git branch, enabling safe parallel execution and instant cleanup of failed attempts.

Provides complete logs and command history, allowing developers to see exactly what agents executed, not just their summaries.

Enables dropping into any agent's terminal to inspect state and take control when agents encounter issues.

Uses standard git commands like `git checkout` to manage agent work, integrating seamlessly with existing developer workflows.

Works with any MCP-compatible agent, model, or infrastructure, avoiding vendor lock-in and supporting diverse setups.