I've been frustrated with LangChain and similar frameworks being impossible to audit, so I built **picoagent** — an ultra-lightweight AI agent that fits in your head. **The core idea:** Instead of guessing which tool to call, it uses **Shannon Entropy** (H(X) = -Σp·log₂(p)) to decide when it's confident enough to act vs. when to ask you for clarification. This alone cuts false positive tool calls by ~40-60% in my tests. **What it does:** - 🔒 Zero-trust sandbox with 18+ regex deny patterns (rm -rf, fork bombs, sudo, reverse shells, path traversal — all blocked by default) - 🧠 Dual-layer memory: numpy vector embeddings + LLM consolidation to MEMORY.md (no Pinecone, no external DB) - ⚡ 8 LLM providers (Anthropic, OpenAI, Groq, DeepSeek, Gemini, vLLM, OpenRouter, custom) - 💬 5 chat channels: Telegram, Discord, Slack, WhatsApp, Email - 🔌 MCP-native (Model Context Protocol), plugin hooks, hot-reloadable Markdown skills - ⏰ Built-in cron scheduler — no Celery, no Redis **The only 2 dependencies:** numpy and websockets. Everything else is Python stdlib. **Where I need help:** - Testing the entropy threshold — does 1.5 bits feel right for your use case or does it ask too often / too rarely? - Edge cases in the security sandbox — what dangerous patterns am I missing? - Real-world multi-agent council testing - Feedback on the skill/plugin system Would love brutal feedback. What's broken, what's missing, what's over-engineered?

picoagents

ai-agent

ai-agent