diff --git a/demos/continuous_batching/agentic_ai/README.md b/demos/continuous_batching/agentic_ai/README.md
index 2c80058c68..cca3bb3746 100644
--- a/demos/continuous_batching/agentic_ai/README.md
+++ b/demos/continuous_batching/agentic_ai/README.md
@@ -488,6 +488,8 @@ To simulate such type of load, we should use a dedicated tool [multi_turn benchm
 ```bash
 git clone -b v0.10.2 https://github.com/vllm-project/vllm
 cd vllm/benchmarks/multi_turn
+wget https://www.gutenberg.org/ebooks/1184.txt.utf-8
+mv 1184.txt.utf-8 pg1184.txt
 pip install -r requirements.txt
 sed -i -e 's/if not os.path.exists(args.model)/if 1 == 0/g' benchmark_serving_multi_turn.py
 # Testing single client scenario, for example with GPU execution