Welcome to Kern

Kern is a lightweight Python framework designed specifically for building AI agents that run efficiently on small language models (1-7B parameters).

While general agent frameworks assume massive cloud LLMs (like GPT-4), Kern is built from the ground up for resource-constrained environments (like running Ollama or llama.cpp on your laptop). It provides structured outputs, automatic JSON repair, tool use, and rich workflow visualization—all tuned to work reliably with smaller models.

Why Kern?

1. Optimized for Small Models (1-7B)

Every design choice in Kern prioritizes running local, small models (such as Llama 3.2 3B, Qwen 2.5 3B, and Phi-4 Mini). Prompts, defaults, and parameters are tuned to maximize the reasoning capabilities of models running locally.

2. Template-Based Structured Output

Small models struggle with abstract JSON Schemas and get confused by deep nesting. Kern uses intuitive fill-in-the-blank templates instead. This mirrors how text-completion models naturally think, dramatically improving structured output accuracy.

3. Automatic JSON Repair

When small models omit closing braces, add trailing commas, or prefix JSON with conversational commentary, Kern's built-in repair engine intercepts the output, fixes the structure, and returns valid JSON.

4. Rich Visualization & Reliability Dashboard

Kern features native Mermaid workflow visualization and a dynamic trace viewer. Build complex multi-agent teams and visualize the execution pipeline, tool calls, and data flow in real time without relying on cloud observability services.

---

Core Capabilities

Getting Started

1. Install Kern:

   1pip install kern-ai

2. Run your first local agent:

   • Build your first agent →
   • Configure model providers →
   • Understand workflows and teams →