# Deep Agent

This repository contains a minimal implementation of a **Deep Agent** inspired by the *Deep Agents from Scratch* course. The agent can:

1. Search the web (DuckDuckGo).
2. Create, read and combine virtual files.
3. Export the virtual files to the real filesystem.
4. Use LangChain + Ollama for LLM inference.

## Setup

```bash
# Install dependencies
pip install -r requirements.txt
```

## Usage

```bash
python deep_agent.py "Write a summary of the latest Python release and store it in summary.txt"
```

The agent will perform the task, create virtual files, and export them to `output_files/`.

## Architecture

- `DeepAgent` class implements the agent logic.
- `search` uses DuckDuckGo HTML API.
- `write_file`, `read_file`, `combine` manage a simple in‑memory virtual filesystem.
- `export_files` writes the virtual files to disk.
- The agent loop uses a JSON‑based action protocol.

## Requirements

- Python 3.10+
- LangChain >= 1.0.0
- LangGraph >= 1.0.0
- langchain_ollama
- langchain_text_splitters
- langchain_chroma
- requests

---

Feel free to extend the agent with more actions or integrate other LLM providers.