Merge pull request #1243 from getlarge/moltnet/00bbac38-2c65-4cf6-8e2e-a70f1a1d1a04/consider-dropping-baked-in-embedding-model-in-favor-of-start

feat(rest-api): drop baked-in embedding model

Author: getlarge

apps/rest-api/Dockerfile

@@ -55,14 +55,13 @@ LABEL org.opencontainers.image.description="MoltNet REST API"
 LABEL org.opencontainers.image.licenses="MIT"
 
 ENV NODE_ENV=production \
-    EMBEDDING_CACHE_DIR=/app/dist/models \
-    EMBEDDING_ALLOW_REMOTE_MODELS=false
+    EMBEDDING_CACHE_DIR=/app/models \
+    EMBEDDING_ALLOW_REMOTE_MODELS=true
 WORKDIR /app
 
 # Runtime tree: app + production node_modules, linux-native.
 # `pnpm deploy` copies only the package's `files` glob (dist/, public/) plus
 # the trimmed node_modules; dev configs and source are excluded automatically.
-# `dist/models/` rides along inside `dist/` (host-built by `download-model`).
 COPY --from=build /out ./
 
 # Drizzle migration SQL — consumed by the migrate entrypoint and Fly.io

apps/rest-api/README.md

@@ -17,8 +17,9 @@ docker compose --env-file .env.local up -d
 
 ### Embedding model (one-time setup)
 
-The API eagerly warms the ONNX embedding pipeline on startup. The model must be
-pre-downloaded locally before running in offline mode.
+The API eagerly warms the ONNX embedding pipeline on startup. By default, the model will be downloaded from Hugging Face at startup time.
+
+For offline development or to avoid repeated downloads, you can pre-download the model:
 
 **Why `apps/rest-api/models/` and not the repo root?**
 pnpm sets the process cwd to `apps/rest-api/` when running the dev script, so

apps/rest-api/package.json

@@ -111,36 +111,9 @@
           "{projectRoot}/dist/migrate.js"
         ]
       },
-      "download-model": {
-        "cache": true,
-        "dependsOn": [
-          "build"
-        ],
-        "executor": "nx:run-commands",
-        "inputs": [
-          {
-            "runtime": "node --version"
-          },
-          "{workspaceRoot}/tools/download-embedding-model.mjs",
-          {
-            "externalDependencies": [
-              "@huggingface/transformers",
-              "onnxruntime-node"
-            ]
-          }
-        ],
-        "options": {
-          "command": "node tools/download-embedding-model.mjs --cache-dir apps/rest-api/dist/models",
-          "cwd": "{workspaceRoot}"
-        },
-        "outputs": [
-          "{projectRoot}/dist/models"
-        ]
-      },
       "docker:build": {
         "dependsOn": [
-          "build:migrate",
-          "download-model"
+          "build:migrate"
         ]
       },
       "e2e": {