ReAct Agent from Scratch using LangChain

This project is a minimal, transparent implementation of a ReAct (Reasoning and Acting) Agent built manually using the LangChain framework.

Instead of relying on high-level abstractions like initialize_agent, this repo walks through how a ReAct agent works under the hood — from formatting the prompt, invoking tools, to parsing LLM responses.

---

Features

• Custom ReAct-style Prompt Template
• Tool execution (e.g., get_text_length)
• Manual ReAct-style reasoning/action loop
• Intermediate step tracking (scratchpad)
• Callback handler for logging LLM input/output

---

Note:

This project uses the mistralai/mistral-7b-instruct model hosted for free via OpenRouter. The model is accessed using the OpenAI-compatible API base