Hello Hailo Team,
We are working on a high-speed, event-driven AI inference system optimized for Raspberry Pi 5 AI HAT Plus. Our architecture goal is:
Minimal C++ core for inference speed (HailoRT)
Event-based AI pipeline using a Python microservice (FastAPI)
Libcamera/Picamera2 for direct camera access (bypassing GStreamer overhead)
Hailo Apps Infra for low-latency AI model execution
Issue
After installing HailoRT 4.20.0, we encounter:
ModuleNotFoundError: No module named 'hailort'
Yet, pip3 list | grep hailort
confirms:
hailort 4.20.0
Finding hailort
on the system:
find /usr -name "hailort"
Returns:
/usr/lib/python3/dist-packages/hailo_platform/drivers/hailort
/usr/share/doc/hailort
Even with PYTHONPATH set:
export PYTHONPATH=/usr/lib/python3/dist-packages:$PYTHONPATH
python3 -c "import hailort; print(hailort.__version__)"
Still fails with βNo module named βhailortββ.
Steps Weβve Tried
-
Reinstalling HailoRT:
pip3 uninstall hailort -y && pip3 install hailort --no-cache-dir
Still fails
-
Activating Hailo SDK Setup (If Required):
cd ~/hailo_sdk_installation/ source setup.sh
Still fails
-
Checking Python Environment:
which python3 && python3 --version && pip3 --version
Only one Python version is installed (
Python 3.11
)
Our Architecture Goals
We are building a high-performance AI pipeline where:
- C++ handles inference for maximum speed.
- Python FastAPI serves events and AI results.
- Picamera2 (Libcamera) directly feeds frames into Hailo models.
- WebSockets provide live AI updates instead of polling.
Given this, our primary concern is how to correctly load and use HailoRT in a mixed C++/Python system.
Key Questions
- Is
hailort
expected to be imported ashailo_platform.hailort
or something else in 4.20.0? - Does HailoRT 4.20 require an additional environment setup?
- Would a direct C++ integration of HailoRT bypass this Python issue entirely?
- Should we be using
hailo-apps-infra
differently for RPi5 AI HAT Plus?
Would appreciate any insights from the community!
Related Post
RPI 4.20 Release Update Discussion