AbstractCore v2.13.15

Added

• Voice and vision provider-availability catalogs: the capability registry now exposes lightweight available_providers() queries for voice and vision backends so server routes can report what is configured without constructing heavy local runtimes.
• Qwen3.6 MTP GGUF catalog entries: added capability metadata for unsloth/Qwen3.6-27B-MTP-GGUF and unsloth/Qwen3.6-35B-A3B-MTP-GGUF, plus explicit GGUF quant selector resolution for model ids such as :Q4_K_M and :UD-Q4_K_M.

Changed

• Canonical server auth naming: the HTTP gateway now uses ABSTRACTCORE_AUTH_TOKEN consistently across runtime config, CLI commands, Python config helpers, docs, and Swagger guidance.
• Swagger auth behavior: /docs keeps the native Swagger Authorize flow, but only advertises the AbstractCore bearer scheme when server auth is actually enabled and validates the token through /acore/auth/validate before storing it client-side.
• Media gateway consistency: image/audio catalog and generation routes now share a cleaner provider/model/base_url override contract and use the current AbstractVoice/AbstractVision capability package floors (abstractvoice>=0.10.3, abstractvision>=0.3.6).
• Apple aggregate profile alignment: abstractcore[all-apple] now matches the current Apple-local dependency stack required by abstractvoice[all-apple], abstractvision[all-apple], mflux, and the newer llama.cpp bindings.
• OpenAI-compatible env surface: the generic OpenAI-compatible provider, registry, config manager, and related docs/tests now consistently use OPENAI_BASE_URL and OPENAI_API_KEY.

Fixed

• all-apple install resolution: repaired the resolver conflict between numpy, Pillow, torch, mflux, and plugin extra floors so pip install -e ".[all-apple]" no longer backtracks across an impossible dependency graph.
• Swagger false authorization state: the browser docs no longer show a misleading authorized state for arbitrary bearer values when server auth is configured.
• Generated media provider metadata: multimodal output routing now preserves explicit voice/vision provider selectors in generated artifacts and resource metadata instead of collapsing them back to the active LLM provider class name.