Hancom-AI Set up

[Vkunduru]

Dear,
I am trying to set up the Hancom-AI features available with the hybrid version. The server connection to the backend is crashing, I have Python 3.14 and Java SDK 26.0.1 . I have installed "opendataloader-pdf[hybrid]". Pyproject.toml is exiting with error when installing opendataloader-pdf[hybrid]. Are there any steps I should re consider? Looking for guidance.

[sueun-dev]

hi, I’d split this into install vs Hancom-AI first. I just tried a clean CPython 3.14.4 venv on macOS arm64 with pip 26.1.1: opendataloader-pdf[hybrid]==2.4.7 installed, opendataloader-pdf-hybrid --no-ocr --device cpu started, and /health returned {"status":"ok"} twice. So Java 26 probably isn’t the first thing I’d chase if the error happens during pip install.

I’d try the smallest clean baseline first:

python -m venv .venv
. .venv/bin/activate
python -m pip install -U pip
python -m pip install -U "opendataloader-pdf[hybrid]==2.4.7"
opendataloader-pdf-hybrid --host 127.0.0.1 --port 5002 --no-ocr --device cpu
curl http://127.0.0.1:5002/health

If that works, remove --no-ocr and then try the Hancom-AI path. If install still fails, the useful line is usually the first package name above the pyproject.toml error, not the final pip summary. For the Java side, JDK 26 is newer than the README baseline, so JDK 17 or 21 LTS is also worth trying if the crash only happens after Python install is clean.

[Vkunduru]

Hi,
So I am running Python 3.12 in a Jupyter Notebook ipykernel. I have used thepip install -U "opendataloader-pdf[hybrid]==2.4.7" command and am receiving this error Building wheels for collected packages: docling-parse Building wheel for docling-parse (pyproject.toml) ... error error: subprocess-exited-with-error. Should I install docling-parse separately, and if so how do I do so? Also the port 5002 seems to be closed, is this an error on my end? The connection seems to "fail to connect".

[sueun-dev]

I tried the Python 3.12 path directly in a clean CPython 3.12.13 venv. pip install -U "opendataloader-pdf[hybrid]==2.4.7" installed cleanly for me, and docling-parse came from a wheel: docling_parse-6.2.0-cp312-cp312-macosx_14_0_arm64.whl, so you normally shouldn’t need to install it separately.

For the 5002 part, that port is only open after the hybrid server finishes startup. In my run it was closed for a few seconds while DocumentConverter initialized, then /health returned {"status":"ok"}.

I’d check the notebook kernel first:

python -c "import sys, platform; print(sys.executable); print(sys.version); print(platform.platform())"
python -m pip --version
python -m pip install -U pip setuptools wheel
python -m pip install -U --only-binary=:all: docling-parse==6.2.0

If --only-binary can’t find a wheel, then it’s probably a platform/Python mismatch, not something fixed by installing docling-parse separately. Also make sure the server is started in a separate terminal/process before the notebook tries to connect to 127.0.0.1:5002.

[Vkunduru]

Hi,
OpenDataloader-pdf[hybrid]==2.4.7 seemed to install fine. I ran the install in a python3.14 venv. However, opendataloader-pdf-hybrid --host 127.0.0.1 --port 5002 --no-ocr --device cpu results in opendataloader-pdf-hybrid: command not found and curl http://127.0.0.1:5002/health results in curl: (7) Failed to connect to 127.0.0.1 port 5002: Connection refused. What should I do in this case?

[Vkunduru]

Hi,
A clean re-install of Python seemed to do the job. I was able to run the function. I had to start a server connection before calling the method as suggested and also mentioned in the details section of the --hybrid parameter command on the documentation page.