python-ai-kit 0.14.3

AI Kit is a CLI meant to augment your IDE agent.

AI Kit

AI Kit is designed for your IDE's agent.

Built to integrate with text editors like Cursor or Windsurf (or any environment with a shell), it extends your agent with search, reasoning, and memory.

Getting Started 🚀

Since AI Kit leverages the most cutting edge models at any given time, you'll need some API keys. Fortunately, the cost of running these models is very low.

Required API Keys:

• GROQ_API_KEY - For lightning-fast thinking with r1-70B (get it here)
• TOGETHER_API_KEY - For deep thinking and reasoning with r1-670B (sign up)
• GEMINI_API_KEY - For smart routing (grab one)

Optional API Keys:

• COHERE_API_KEY - For reranking search results (get access)

Drop these in your .env file at your project root.

Quick Setup

# 1. Initialize AI Kit (creates necessary dirs and system prompts)
ai-kit init

# 2. Add your API keys to .env
GROQ_API_KEY=your_key_here
TOGETHER_API_KEY=your_key_here
GEMINI_API_KEY=your_key_here
COHERE_API_KEY=your_key_here  # Optional

# 3. Verify everything's working
ai-kit help

That's it! Run ai-kit help anytime to check your setup status, or ai-kit status to view API keys.

The "Brain" 🧠

graph TD
    User([User])
    Exec["Executioner LLM<br/>1) Shell<br/>2) Read/Write<br/>3) Local Grep"]
    Router["Router (Gemini)"]
    Think["Quick Think (Groq R1-70B)"]
    DeepThink["Deep Think (Together R1)"]
    Prompts[(User Prompts)]
    System[(System Prompts)]
    
    User <--> Exec
    Exec --> Router
    
    subgraph Brain
        Router -->|simple| Exec
        Router -->|think| Think
        Router -->|deep think| DeepThink
        Think --> Prompts
        Think --> System
        Think -->|thought stream| Exec
        DeepThink --> Prompts
        DeepThink --> System
        DeepThink -->|thought stream| Exec
    end

The thinking system (brain) has three main components:

1. The IDE Agent - This is the built in agent in your IDE. I'd recommend using Cursor + Claude.
2. The Router - Uses Gemini to route queries to the appropriate reasoning LLM.
3. The Thinking LLM - Injects thinking tokens from R1 into the agent's context.

Tools

The agent has access to a few extra tools:

• search - Search the web for information
• fetch - Fetch a URL and return the content
• crawl (beta) - Crawl a website and save the content

Principles 🎯

• Local first, for full control
• Hardcode as little as possible, instead use composable patterns and leverage agency
• Runtime first, prepare as little as possible, give the agent tools instead
• Use a non-reasoning model (like Claude-3.5-sonnet) for tool calls and edits and rely on reasoning models for planning and orchestration