Build Agentic workflows, RAG pipelines, with rich AI model and tool support on one collaborative workspace. Deploy on cloud, VPC, or self-hosted, so teams move from prototype to production without rebuilding the stack.
Python agentagentic-aiagentic-frameworkagentic-workflowaiautomationclaudedeepseekgenaigptllmlow-codemcpnextjsno-codeopenaipythonskillsworkflow
#!/usr/bin/env sh
set -eu
script_dir=$(CDPATH='' cd -- "$(dirname -- "$0")" && pwd)
project_dir=$(CDPATH='' cd -- "$script_dir/../../../.." && pwd)
container_name="dify-agent-runtime-backend-integration-$$"
image="${DIFY_AGENT_TEST_LOCAL_SANDBOX_IMAGE:?set DIFY_AGENT_TEST_LOCAL_SANDBOX_IMAGE to a Local Sandbox image built from the same commit}"
token="${DIFY_AGENT_TEST_LOCAL_SHELLCTL_AUTH_TOKEN:-runtime-backend-integration}"
cleanup() {
docker rm -f "$container_name" >/dev/null 2>&1 || true
}
trap cleanup EXIT INT TERM
docker run --detach --rm \
--name "$container_name" \
--env SHELLCTL_ENABLE_PATH_ISOLATION=true \
--env SHELLCTL_AUTH_TOKEN="$token" \
--publish 127.0.0.1::5004 \
"$image" >/dev/null
published_address=$(docker port "$container_name" 5004/tcp | head -n 1)
endpoint="http://$published_address"
attempt=0
until curl --fail --silent "$endpoint/healthz" >/dev/null; do
attempt=$((attempt + 1))
if [ "$attempt" -ge 100 ]; then
docker logs "$container_name"
exit 1
fi
sleep 0.1
done
cd "$project_dir"
NO_PROXY=127.0.0.1,localhost \
DIFY_AGENT_TEST_LOCAL_SHELLCTL_ENDPOINT="$endpoint" \
DIFY_AGENT_TEST_LOCAL_SHELLCTL_AUTH_TOKEN="$token" \
pdm run pytest --import-mode=importlib \
tests/integration/dify_agent/runtime_backend/test_runtime_backend_lifecycle.py \
-k local -q -rs "$@"