The agent that grows with you
Python aiai-agentai-agentsanthropicchatgptclaudeclaude-codecodexhermeshermes-agentllmnous-researchopenai
fix(docker): new hosted agents no longer rebuild their venv at first boot to get fal
IAvecilla0c74cc682dc8
diff --git a/Dockerfile b/Dockerfileindex 779964df82f..0d5d0ead27a 100644--- a/Dockerfile+++ b/Dockerfile@@ -272,6 +272,11 @@ FROM runtime_base AS python_deps# is baked so hosted/immutable images can enable the adapter without writing# the sealed venv.#+# [fal] is baked because the image tool's availability check imports+# fal_client whenever a FAL backend resolves, which the Nous-managed gateway+# always does on hosted agents. Lazy-installing it at the gateway's first boot+# builds a whole new venv generation on the volume (about 90s).+## --record-selection writes these extras to the store's facts.json as the venv# selection, as a bundle build does. Without it, the first opt-in extra a# container installs on use (FAL, a plugin's deps) records a selection of@@ -291,7 +296,7 @@ RUN python3 -m pm.build_env --source /opt/hermes --python /usr/local/bin/python3--cache /opt/hermes/uv-cache \--out /opt/hermes/.venv --no-install-project --sealed --record-selection \--extra all --extra messaging --extra otlp --extra anthropic --extra bedrock \- --extra azure-identity --extra matrix --extra google-chat+ --extra azure-identity --extra matrix --extra google-chat --extra fal# Icons render on the runtime environment: Pillow and resvg-py are core# dependencies. A stage of its own so the frontend stage keeps building its