Todos¶

Self-registration stays open with sso and ldap off: GET /api/config reports features.enable_signup: true while an administrator exists, so any visitor can create a user account (DEFAULT_USER_ROLE=user in templates/env.j2). ENABLE_SIGNUP keeps its upstream default true, and ENABLE_PERSISTENT_CONFIG=false reads it again on every start. Options: render ENABLE_SIGNUP=false in templates/env.j2, or set DEFAULT_USER_ROLE=pending so a self-registered account waits for an administrator.
tasks/utils/mcp_grants.yml passes OPENWEBUI_API_BOT_USER.password and the tool server connections with their bearer tokens as -e NAME=<value> argv elements of container exec, so a -vv deploy log prints them; forward each as -e NAME plus environment: like tasks/utils/admin.yml does.
Logo and title stay Open WebUI’s own (meta/services.yml design entry, operator decision of 2026-10-13): clause 4 of the Open WebUI license at v0.11.0 (LICENSE) allows replacing its branding only for deployments of up to 50 users in 30 days, with written permission from the licensor, or under an enterprise license. Options: obtain written permission or an enterprise license, then switch logo and title on in the design entry.
The first boot serves nothing until the embedding model is on disk: uvicorn binds only after the lifespan startup, which downloads sentence-transformers/all-MiniLM-L6-v2 (30 files) from Hugging Face into the data volume. Measured on a slow network (2026-10-14, UTC): container created 08:04:17, Waiting for application startup 08:06:27, model loaded 08:14:31, so the proxy answered 502 for eight minutes and the readiness wait of the test stage (08:11:07 to 08:13:47, 30 attempts) failed the deploy. The image healthcheck has no start period, so sys-ctl-rpr-container-soft restarted the still downloading container as unhealthy at 08:05:59 and the download began again. Options: embed through the LiteLLM gateway (RAG_EMBEDDING_ENGINE=openai) when litellm is on, give the service a healthcheck start_period that covers the download, or fetch the model in a task before the readiness wait.