fak

What it does

fak (Fused Agent Kernel) is a lightweight Go binary designed to power AI agent loops with efficiency and flexibility. It wraps Claude, Code, Codex, and Cursor integrations while enabling developers to run local GGUF models. The tool optimizes long-running sessions through cache-efficient KV management and prompt caching, reducing latency and costs for extended agent interactions.

Key Features

• Single Go binary for streamlined deployment and execution
• Multi-provider support including Anthropic, OpenAI, and local models
• KV-cache optimization for efficient long-session handling
• Local GGUF model inference without external dependencies
• Intelligent tool call adjudication and routing
• Prompt caching to reduce redundant processing
• MCP (Model Context Protocol) compatibility
• Per-call work routing for granular control

How to set up

Clone the repository from GitHub and build the Go binary for your platform. Install dependencies as specified in the documentation. Configure your API keys for desired providers (Anthropic, OpenAI) or point to local GGUF models. Run the binary with appropriate configuration flags to initialize your agent loop, adjusting cache settings and model routing based on your use case.