Use uv venv for embedding, pre-download model in Docker build

- Use /app/venv with uv instead of system python/pip
- Pre-download all-MiniLM-L6-v2 model during Docker build

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Dockerfile

@@ -8,10 +8,11 @@ RUN npm run build
 
 # Stage 2: Runtime
 FROM alpine:3.21
-RUN apk add --no-cache ca-certificates curl bash python3 py3-pip
+RUN apk add --no-cache ca-certificates curl bash
 RUN curl -LsSf https://astral.sh/uv/install.sh | sh
 ENV PATH="/root/.local/bin:$PATH"
-RUN pip3 install --break-system-packages sentence-transformers
+RUN uv venv /app/venv && uv pip install --python /app/venv/bin/python sentence-transformers
+RUN /app/venv/bin/python -c "from sentence_transformers import SentenceTransformer; SentenceTransformer('all-MiniLM-L6-v2')"
 RUN mkdir -p /app/data/workspaces
 WORKDIR /app
 COPY target/aarch64-unknown-linux-musl/release/tori .

src/kb.rs

@@ -101,7 +101,7 @@ impl KbManager {
 async fn compute_embeddings(texts: &[String]) -> Result<Vec<Vec<f32>>> {
     let input = serde_json::json!({ "texts": texts });
 
-    let mut child = tokio::process::Command::new("python3")
+    let mut child = tokio::process::Command::new("/app/venv/bin/python")
         .arg("/app/scripts/embed.py")
         .stdin(Stdio::piped())
         .stdout(Stdio::piped())