From cc61211aed49a777b517945fb34ed7ee4d7b99df Mon Sep 17 00:00:00 2001
From: Quoc-Anh Nguyen <9426721+imcvampire@users.noreply.github.com>
Date: Sat, 13 Jun 2026 14:08:46 +0300
Subject: [PATCH] fly: Enhance Dockerfile with local embeddings installation

Added installation of optional peer dependency for local embeddings and preloaded model weights during Docker build.
---
 deploy/fly/Dockerfile | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/deploy/fly/Dockerfile b/deploy/fly/Dockerfile
index 89257f937..9942be17e 100644
--- a/deploy/fly/Dockerfile
+++ b/deploy/fly/Dockerfile
@@ -25,6 +25,13 @@ RUN printf '{"name":"agentmemory-deploy","version":"1.0.0","private":true,"overr
  && npm install "@agentmemory/agentmemory@${AGENTMEMORY_VERSION}" --omit=optional --no-fund --no-audit \
  && ln -s /opt/agentmemory/node_modules/.bin/agentmemory /usr/local/bin/agentmemory
 
+# Local embeddings (Xenova/all-MiniLM-L6-v2, 384-dim). The optional peer dep
+# is omitted by --omit=optional above, so install it explicitly and preload
+# the model weights at build time (no runtime HuggingFace download on cold
+# start, no volume cache needed).
+RUN npm install @xenova/transformers --no-fund --no-audit \
+ && node -e "const {pipeline}=require('@xenova/transformers');pipeline('feature-extraction','Xenova/all-MiniLM-L6-v2').then(()=>console.log('agentmemory: embedding model cached'))"
+
 ENV AGENTMEMORY_III_VERSION=${III_VERSION} \
     TINI_SUBREAPER=1