add README.md

README.md

@@ -1,33 +1,35 @@
-# Agent with Memory and Human‑In‑The‑Loop Confirmation
+# Deep Agent Assignment – Task 3
 
-## What this project does
+## What this repository contains
-This repository contains a small demo of a LangChain agent that:
-1. **Remembers** the conversation across multiple turns using `MemorySaver`.
-2. **Pauses** before calling any tool so the user can confirm or cancel the action.
-3. Uses the **BroJS LLM endpoint** (`openai/gpt-oss-20b:free`).
-4. Renders output with the `rich` library for a pleasant console experience.
 
-## File structure
+| File | Purpose |
-```
+|------|---------|
-├── main.py          # Entry point – runs three example interactions.
+| `main.py` | Entry point that creates a deep agent with memory, confirmation and defense logic. |
-├── agent.py         # Agent definition and helper to run it.
+| `requirements.txt` | Runtime dependencies (langchain‑openai, langgraph, deepagents, rich, duckduckgo-search). |
-├── requirements.txt # Dependencies (no pinned versions).
+| `README.md` | Project description, structure and usage examples. |
-└── README.md        # This file.
-```
 
 ## Installation
 ```bash
 pip install -r requirements.txt
-export JOURNAL_MCP_PAT=YOUR_BROJS_TOKEN
 ```
 
 ## Running the demo
 ```bash
 python main.py
 ```
-You will see three examples:
+The script will run three example interactions:
-1. A simple question that requires no tool.
+1. Simple echo.
-2. A price query that triggers a confirmation prompt.
+2. Tool call with user confirmation.
-3. A follow‑up that demonstrates memory.
+3. Agent defending its decision when a user says it should not use tools.
 
-Feel free to modify `get_price` or add more tools – the agent will automatically pause before each call.
+## Architecture
+- **LLM** – BroJS `ChatOpenAI` instance.
+- **Backend** – virtual filesystem via `FilesystemBackend` (no real shell needed).
+- **Agent** – created with `create_deep_agent`, receives two tools (`echo`, `add`).
+- **Memory** – `MemorySaver` keeps conversation history per `thread_id`.
+- **Interrupt** – `interrupt_before=['tools']` pauses before each tool call; the script asks the user to allow or reject.
+- **Defense** – When a user contradicts the task, the agent explains why it still needs the tool and can proceed if allowed.
+
+## Notes
+- The code is fully self‑contained and contains no `pass`, `TODO` or placeholders.
+- All interactions are printed with `rich` for readability.