dsh-local-ocr-runtime
v0.2.0
Published
Managed local PaddleOCR runtime for the DeepSeek Harness local OCR plugin
Readme
dsh-local-ocr-runtime
The runtime owns the Python virtual environment, PaddleOCR model cache, service logs, and process state. Nothing is written into the repository by default.
npx dsh-local-ocr-runtime doctor
npx dsh-local-ocr-runtime setup --yes
npx dsh-local-ocr-runtime start
npx dsh-local-ocr-runtime status --json
npx dsh-local-ocr-runtime stopsetup prints the locked PaddleOCR 3.7.0 / PaddlePaddle 3.3.1 model source,
model family, estimated size, cache path, and local-only privacy behavior before asking for consent. --yes is the
explicit consent flag for unattended setup. --skip-model-download records
consent but leaves models deferred; start will then return
OCR_MODEL_NOT_READY until setup is rerun.
The npm package carries the verified local_ocr_service source under
service/, so a published install does not require a source checkout or a
separately published Python package. The command also accepts --service-dir
for a repository development checkout. The service always binds to
http://127.0.0.1:<port> and rejects path-bearing or non-loopback URLs.
Stable error codes include OCR_RUNTIME_NOT_INSTALLED,
OCR_RUNTIME_NOT_RUNNING, OCR_MODEL_NOT_READY, OCR_VERSION_MISMATCH, and
OCR_RUNTIME_PROCESS_CONFLICT. Error output includes a command that can fix
the condition. stop verifies the recorded PID command line before sending a
signal, so it will not terminate an unrelated process that reused a stale PID.
doctor also warns about the V2 preview profile patch that forced
deepseek-local-ocr/deepseek-v4-flash. Run the suggested
scripts/install-plugin.ps1 migration and restart Harness; the current bundle
never chooses a provider or model.
Use this Runtime lifecycle consistently. A repository checkout's
scripts/start-ocr-service.ps1 launches a separate development service and is
not owned by this CLI; do not run both against the same 127.0.0.1 port.
