Core Features
Strong Type Contracts
JSON Schema bidirectional validation for inputs and outputs, ensuring reliable structured output.
Control/Data Separation
Meta control plane for routing decisions, data plane for business logic and auditing.
Module Tiers
exec / decision / exploration - three tiers with different constraints for different scenarios.
Explainable Output
Mandatory confidence + rationale output, every call has confidence score and reasoning.
Multi-LLM Support
OpenAI / Anthropic / MiniMax / Ollama - one module, run anywhere.
MCP Integration
Native support for Claude Desktop, Cursor and other AI tools, seamless integration.
Quick Start
# Zero-install run
npx cogn run code-reviewer --args "your code" --pretty
# Or install globally
npm install -g cogn
cog run code-reviewer --args "your code" --pretty