KISS AI is a Python framework designed to simplify the development of AI agents and evolutionary systems. It addresses the complexity found in many existing AI agent frameworks by prioritizing a "Keep It Simple, Stupid" approach, enabling developers to build and orchestrate multi-agent systems with standard Python code. The framework is beneficial for researchers and power users looking for a streamlined yet powerful toolset for AI agent development, optimization, and experimentation.

How It Works

KISS treats AI agents as functions, allowing for straightforward multi-agent orchestration through function composition in plain Python, eliminating the need for complex message buses or state machines. It features a RelentlessCodingAgent for long-running tasks with auto-continuation, a RepoOptimizer for iterative code optimization based on specified metrics, and GEPA for genetic-Pareto prompt optimization. This design prioritizes clarity, extensibility, and direct expression of intent, integrating advanced AI techniques into a simple, cohesive framework.

Quick Start & Requirements

• Primary install / run command:
  • Install as a library: pip install kiss-agent-framework
  • Launch web assistant: python -m kiss.agents.assistant.assistant
  • Development install: curl -LsSf https://raw.githubusercontent.com/ksenxx/kiss_ai/refs/heads/main/install.sh | sh
• Non-default prerequisites and dependencies: API keys for LLM providers (Anthropic, Gemini, OpenAI, OpenRouter, Together AI) are required. Python 3.13 is recommended for development. Docker is supported for isolated execution.
• Links: PyPI

Highlighted Details

• Multi-Agent Orchestration: Agents are composed as Python functions, enabling simple orchestration via standard Python code.
• Relentless Coding Agent: A single-agent system with smart auto-continuation for long-running coding tasks, supporting Docker execution.
• Repo Optimizer: Iteratively optimizes code within a project repository for metrics like running time and cost, using the RelentlessCodingAgent.
• GEPA Prompt Optimization: Implements a Genetic-Pareto prompt optimization framework for evolving prompts using natural language reflection.
• Trajectory Saving and Visualization: Automatically saves agent execution histories and provides a web-based visualizer for reviewing agent conversations, tool calls, and token usage.
• Model Agnostic: Supports multiple LLM providers (OpenAI, Anthropic, Gemini, Together AI, OpenRouter) and native function calling.

Maintenance & Community

The primary author is Koushik Sen (ksen@berkeley.edu). No specific community channels (e.g., Discord, Slack) or active sponsorship details are provided in the README.

Licensing & Compatibility

• License type: Apache 2.0.
• Compatibility notes: The Apache 2.0 license is permissive and generally compatible with commercial use and closed-source linking.

Limitations & Caveats

Some Gemini models are noted as having "preview, unreliable function calling." The KISSEvolve framework, while kept for historical reasons, was superseded by the repo_optimizer for efficiency and simplicity, suggesting it may be less actively maintained. The framework requires API keys for LLM providers, which may incur operational costs.