diff --git a/cluster/launch_autobot.sh b/cluster/launch_autobot.sh
new file mode 100755
index 0000000..3b73bb0
--- /dev/null
+++ b/cluster/launch_autobot.sh
@@ -0,0 +1,54 @@
+#!/bin/bash
+
+# We want an option to take GPU_TYPE and NUM_GPUS as arguments. They should be flags.
+
+# Parse the arguments
+while getopts ":t:n:" opt; do
+  case $opt in
+    t) GPU_TYPE=$OPTARG ;;
+    n) NUM_GPUS=$OPTARG ;;
+    \?) echo "Invalid option -$OPTARG" >&2; exit 1 ;;
+  esac
+done
+
+# Shift the arguments so that $@ contains only the command to run
+shift $((OPTIND-1))
+
+# Default values
+GPU_TYPE=${GPU_TYPE:-rtx3090}
+NUM_GPUS=${NUM_GPUS:-1}
+
+
+OUTPUT=$(python -m rpad.core.autobot available ${GPU_TYPE} ${NUM_GPUS} --quiet --local)
+
+# Split the output on the colon
+NODE=${OUTPUT%%:*}
+GPU_INDICES=${OUTPUT#*:}
+COMMAND=$@
+
+# Print the results (optional)
+echo -e "######################################"
+echo -e "Launching job:"
+echo -e "  Node:\t\t${NODE}"
+echo -e "  GPU type:\t${GPU_TYPE}"
+echo -e "  GPU indices:\t${GPU_INDICES}"
+echo -e "  Command:\t${COMMAND}"
+echo -e "######################################"
+
+# For some reason, the -c option causes an error "option requires an argument", but it still works.
+ssh -t ${NODE} bash -ic "
+# Setup env.
+cd ~/code/rpad/python_ml_project_template
+source ~/miniconda3/bin/activate
+conda activate python_ml_project_template
+
+# Now, treat the rest of the command line arguments as the command to run.
+CUDA_VISIBLE_DEVICES=$GPU_INDICES $COMMAND
+
+echo '######################################'
+echo 'Job finished.'
+echo '######################################'
+
+# Deactivate the conda environment.
+conda deactivate
+"