AnythingLLM bundles its own inference engine as Phoenix and RAGFlow harden toward production defaults.

The week in ai-ml-platforms

Four products updated, and the pattern is operational maturity rather than new model capability. Arize Phoenix flipped authentication on by default, RAGFlow declared a final 1.0 candidate, and LiteLLM kept stacking breaking changes into its next feature line.

The lone spark is AnythingLLM, which bundles an inference engine into its Docker image. That collapses a two-tool local setup into one container, the most structural change of the week.

Leaders

AnythingLLM v1.19.0 puts AMD's Lemonade inference server inside the Docker container, so a single docker run with ENABLEBUILTIN_LLM=true runs models without Ollama or LM Studio. It ends a two-release stabilization stretch with a structural change.

Arize Phoenix 21.0 turns auth on by default without needing PHOENIX_SECRET and moves model cost data to a manifest fetched at runtime. A companion TypeScript client release adds DECISION and AGENT annotator kinds.

RAGFlow v1.0.0-rc3 adds a decision model that judges whether retrieved evidence is sufficient, the last stop before 1.0.

Wildcards

LiteLLM is running two lines at once: the 1.109 dev build carries 704 PRs and 15 action-required items, while stable branches such as 1.106.5 get backports like leaner failure logs.

Themes that compounded

• Decision and agent concepts appeared in both RAGFlow's decision model and Phoenix's DECISION and AGENT annotator kinds.
• Secure-by-default and less-data-retained changes landed in Phoenix (auth) and LiteLLM (leaner failure logs).
• Self-hosting got easier or stricter in AnythingLLM's single-container path and Phoenix's default auth.

Watch this week

Watch for a RAGFlow 1.0 stable tag following rc3, and for whether LiteLLM ships further 1.109 dev builds that add to its action-required list. Phoenix's default-auth change may also draw follow-up patch releases.