picoagents – ultra-lightweight AI agent framework with entropy-based routing

Summary: picoagents is a minimal AI agent framework using Shannon Entropy to decide when to act or ask for clarification, reducing false tool calls by 40-60%. It features a zero-trust sandbox, dual-layer memory without external databases, and supports multiple LLM providers and chat channels with only two dependencies.

What it does

picoagents routes actions based on entropy confidence, employs an 18+ pattern zero-trust sandbox, and manages memory via numpy embeddings and LLM consolidation. It integrates eight LLM providers, five chat channels, plugin hooks, hot-reloadable Markdown skills, and a built-in cron scheduler without external services.

Who it's for

Developers seeking a lightweight, auditable AI agent framework with customizable security and multi-provider LLM support.

Why it matters

It addresses the lack of auditability and excessive false positives in existing AI agent frameworks by combining entropy-based decision-making with strict sandboxing and minimal dependencies.