fix: skip model-catalog loading in proxy_app to avoid cpu-basic hang

fix: handle missing upstream in models_response, add env to skip litellm remote startup

fix: skip huge catalog, use static config only, skip proxy_app startup render

force rebuild: bust Docker cache v2

force rebuild: bust Docker cache

load all 3 probe results: chat_usable, image_usable, vision_usable

add /admin/discover/results endpoint

remove broken GitHub fetch, all scripts updated

full repo sync

update all scripts: /admin/discover, fix verify base_url

add 5min cooldown between probes, vision gets more retries

remove usable-models cache TTL, always fetch fresh from dataset

add --no-filter to image and vision probes for full coverage

add chat+image+vision probes to /admin/verify

touch proxy_app.py to bust Docker COPY cache

fix: unbuffered python output for real-time probe logs

fix: add early log writes to debug probe startup

fix: stream probe logs line-by-line for real-time viewing

fix: use public URL for verify probe instead of internal port

fix: capture probe output to log file, show upload errors

add /admin/logs endpoint to read probe log in real time

fix: use asyncio subprocess to avoid blocking event loop

fetch usable-models from HF dataset, upload after probe

add /admin/verify endpoint for daily usable-models probe

Make image model checks less brittle

Track image usable models

Refresh model catalog metadata

Track verified usable models

Expose model catalog metadata

Hide image-only routes from chat models

Fix image generation bridge

Start public LiteLLM template history