Universal autonomous agent framework with ReAct loop, multi-provider LLM routing, reasoning graph, and MCP integration, domain-agnostic for building specialized AI agents.

Code snippets for my 2025 lecture at Columbia University, an introduction to AI Agents

Minimal local-Qwen coding agent in a single Python file. ReAct loop, tool dispatch, atomic tools, no framework. Educational.

Quality-gated autonomous mission completion loop for Claude Code and Codex (plan, execute, review, score, iterate).

A pure Python implementation of ReAct agent without using any frameworks like LangChain. It follows the standard ReAct loop of Thought, Action, PAUSE, and Observation. The agent utilizes multiple tools, including Calculator, Wikipedia, Web Search, and Weather. A web UI is also provided using Streamlit.

An interactive AI Agent built with Streamlit that evaluates cat food quality using a ReAct decision loop. Powered by Llama 3.3, it combines real-time DuckDuckGo web searches with custom Python tools to calculate Dry Matter, NFE, and Ca:P ratios. It provides professional veterinary verdicts based on user-selectable FEDIAF or AAFCO standards.

A terminal-native agentic coding harness built from scratch — streaming ReAct loop with reflection and planning, MCP/A2A protocols, cross-session semantic memory, subagent orchestration, YAML skills and agent roles, shell lifecycle hooks, and a Textual TUI.

An autonomous AI agent using Groq's LPU for ultra-fast ReAct loops. Features real-time web search via SerpAPI, safe mathematical evaluation, and local document parsing.

A modular terminal AI agent capable of autonomous code execution, context propagation, and secure OS sandboxing using Gemini and local Ollama engines.

🧠 Local AI coding agent that autonomously finds bugs, writes fixes, self-reviews, and validates with tests. Powered by Ollama. 100% offline. Zero API costs.

Multi-Agent Orchestration System — 17 AI agents coordinating through a ReAct loop to build complete software projects from a single goal

🛡️ An autonomous cybersecurity agent powered by Gemma 3 that automates the network penetration testing lifecycle. ThreatScope uses a ReAct loop and RAG to discover active hosts, triage vulnerabilities via Nmap/Vulners, and generate prioritized, human-readable remediation reports.

WhatsApp channel for Marvin — knowledge-grounded AI assistant with MCP tool access, Milvus RAG, politeness system, and web search fallback.

Local first agentic AI assistant with real-time web search, tool calling, and streaming responses, powered by llama.cpp and your own hardware.