This project explores how to build LLM-powered autonomous agents that can reason, plan, and act through tools to solve complex tasks. Inspired by the Agentic paradigm in modern AI systems, it demonstrates how you can orchestrate tool use, memory, planning, and multi-step reasoning with large language models (LLMs).
LLM agents combine:
- Reasoning (via LLMs like GPT-4, Claude, or Mixtral)
- Tools (like search, calculators, and custom APIs)
- Memory (for short/long-term context retention)
- Planning (breaking tasks into sub-tasks)
- Autonomy (looping through steps until a goal is reached)
This repo shows you how to wire all these together.
-
🪄 Task Decomposition & Planning
Breaks complex instructions into manageable steps. -
🧰 Tool Execution
Supports tool use (e.g., web search, calculator, file reading) from within LLM prompts. -
💭 Multi-step Reasoning Loop
The agent thinks, acts, observes, and reflects repeatedly until a task is completed. -
💾 Memory Handling
Incorporates a basic memory structure to store and recall prior steps or observations. -
🧪 Modular Components
Easily plug in different LLMs, tools, or memory systems.
- Python 3.10+
- LangChain or custom framework
- OpenAI, Anthropic, or open LLMs
- Tool abstractions (Search, Math, File I/O, etc.)
git clone https://github.com/gr8nishan/llm_agents.git
cd llm_agents