Both use an allowlist so only the files each Dockerfile copies are sent
to the build context. For caddy this keeps the TLS certificates and keys
in data/ and config/ out of the context; for api it skips training code,
datasets, saved models and caches.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
- API: read the upload size limit from MAX_UPLOAD_MEGABYTES (default 15)
and keep MAX_UPLOAD_BYTES as the byte value used by /predict
- Caddy: read the ACME email and request body limit from CADDY_EMAIL and
CADDY_MAX_BODY (default 16MB, ~1 MB above the API limit)
- compose: pass the new variables to caddy_proxy and cnn_api
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
marvinandClaude Sonnet 5