diff --git a/commands.sh b/commands.sh
index 1d11ad1..46465e1 100644
--- a/commands.sh
+++ b/commands.sh
@@ -23,6 +23,7 @@ conda activate autog-cpu
 # -------------- install other dependencies int autog-cpu environment
 pip install codetiming humanfriendly sentence_transformers==3.3.0 nltk==3.9.1 torchdata==0.7 torchtext==0.16.0
 pip install transformers==4.44.2
+pip install llama-index llama-index-llms-bedrock
 # for Graphviz
 sudo apt-get install graphviz
 # for nltk data
@@ -70,10 +71,13 @@ bash scripts/download.sh
 #   - input the overall prompt+new contents into an LLM (e.g., Claud Sonet 3.5+) API or Web console to get answer.
 #   - copy the answer (in RAW format!!!!!) and save to a file, named type.txt under the mag/ folder.
 
-# 2. run autog, and explanation of these arguments
+# 2. Option 1: run autog (manually call LLM), and explanation of these arguments
 #   mag             -> name of the --dataset argument
 #   /data/datasets  -> path of the folder that store data, .e.g, /data/datasets
 #   autog-s         -> the method to run the model
 #   type.txt        -> the name of file to save analysis results from LLM. Since we manually create this, this file is not used.
 #   venue           -> Name of the task to fit the solution
 python -m main.autog mag ./data/datasets/mag autog-s type.txt venue
+
+# 2. Option 2: run autog2 (automatically interact with LLM for single or multiple rounds)
+python -m main.autog2 mag anthropic.claude-3-sonnet-20240229-v1:0 ./data/datasets autog-s venue
diff --git a/main/autog2.py b/main/autog2.py
index f989d14..cdf31d5 100644
--- a/main/autog2.py
+++ b/main/autog2.py
@@ -3,7 +3,7 @@
 import typer
 import numpy as np
 from rich import traceback
-from models.autog.agent import AutoG_Agent
+from models.autog.agent_old import AutoG_Agent
 from prompts.task import get_task_description
 from utils.misc import seed_everything
 from utils.data.rdb import load_dbb_dataset_from_cfg_path_no_name
diff --git a/scripts/download.sh b/scripts/download.sh
index 796358b..c688bc8 100644
--- a/scripts/download.sh
+++ b/scripts/download.sh
@@ -1,6 +1,6 @@
 ## IEEE-CIS RR Movielens Outbrain MAG AVS diginetica ESCI Stackexchange
 
-dataset_path="/data/datasets"
+dataset_path="./data/datasets"
 
 mkdir -p $dataset_path