22 lines
1.2 KiB
Markdown
22 lines
1.2 KiB
Markdown
# DeepAgent Implementation
|
||
|
||
This repository contains a minimal **DeepAgent** that can search the web and create virtual files. The agent uses OpenRouter for LLM calls, DuckDuckGo for web searches, and `deepagents` to manage a hybrid filesystem + shell backend.
|
||
|
||
## Features
|
||
- **Web search**: Uses DuckDuckGo API to fetch up to 5 results per query.
|
||
- **Virtual file creation**: The agent writes files into a temporary workspace (`./workspace`).
|
||
- **Real‑world file export**: After completing the task, the virtual files are copied to the real filesystem under `./workspace`.
|
||
|
||
## How it works
|
||
1. The agent is instantiated with an OpenRouter LLM and a single web search tool.
|
||
2. A backend combines a local shell environment (for executing commands) and a simple in‑memory filesystem.
|
||
3. When the user asks for information, the agent calls `web_search`, receives the results, and stores them in `results.txt` inside the virtual workspace.
|
||
4. The script then copies that file to the real working directory so it can be committed to Git later.
|
||
|
||
## Running locally
|
||
```bash
|
||
pip install -r requirements.txt
|
||
python main.py
|
||
```
|
||
The program will print search results and create a `results.txt` file in the current directory.
|