What
`src/asi_build/deployment/universal_huggingface_deployer.py:402`:
api_url = f\"http://localhost:8001/pipeline/{deployment_id}\"
The "universal" deployer hardcodes a local URL with no env override. Same file is what gets used to deploy to remote CUDO/HF endpoints — but as soon as the inference service moves off the box, this URL won't reach it.
Why this matters
- The Memgraph URL (`bolt://localhost:7687`) at least has `os.getenv("MEMGRAPH_URL", ...)` as a fallback. This call site doesn't.
- Anyone wiring up a multi-host deploy will hit a silent "why are predictions hanging" rabbit hole.
Fix
api_base = os.environ.get(\"ASI_BUILD_PIPELINE_URL\", \"http://localhost:8001\")
api_url = f\"{api_base}/pipeline/{deployment_id}\"
And document the env var in `configs/default.yaml` and `CONTRIBUTING.md` next to the existing `MEMGRAPH_URL` docs.
What
`src/asi_build/deployment/universal_huggingface_deployer.py:402`:
The "universal" deployer hardcodes a local URL with no env override. Same file is what gets used to deploy to remote CUDO/HF endpoints — but as soon as the inference service moves off the box, this URL won't reach it.
Why this matters
Fix
And document the env var in `configs/default.yaml` and `CONTRIBUTING.md` next to the existing `MEMGRAPH_URL` docs.