You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Model registry/catalog: expand the curated vision_model_capabilities.json inventory with more current official or reputable vision model families and cross-platform download variants, tighten provider provenance, remove adapter-only repos from the standalone model catalog, and expose richer model notes / task_specs through the CLI catalog surfaces.
Apple/MLX semantics: treat mlx as a download target rather than a runnable backend, fail fast on generic mlx provider/model routing across CLI, playground, and AbstractCore, and limit shipped MFLUX presets/docs to the families the backend actually supports today.
GGUF / local-runtime routing: improve stable-diffusion.cpp bundle resolution for curated multi-artifact models and legacy Qwen GGUF repo ids, resolve curated sdcpp model keys consistently in both one-shot CLI and REPL paths, and make AbstractCore request-scoped sdcpp routing explicit while canonicalizing MFLUX aliases for backend reuse.
Docs/policy/tests: add the ADR set and planned backlog items for curated catalog ownership, adapter-aware model graphs, and the long-term native MLX direction; refresh README/reference/FAQ/agent docs; regenerate llms-full.txt; and expand regression coverage for catalog aliases, GLM defaults, MFLUX routing, and local backend bundle selection.