Remove Pixabay from image pipeline — replace with Pexels/Unsplash

fix: content-hash dedup with _MAX_STOCK_SHARING=4 — eliminate duplicate photos in Oslo, Osaka, Marrakech, Hanoi, Prague, Madrid caches

fix: content-hash dedup for Pixabay, Pexels, and Unsplash — zero duplicate photos across all stock providers

feat: add Pexels as tier 7 in image enrichment chain

fix: prefer Wikimedia images over Pixabay stock photos on re-check

feat: fast offline translations via deep-translator with parallel HTTP

v0.1.22 — Move cache files to .cache/ directory

Revert "v0.1.21 — Phase 2 polish: .gitignore cleanup, docstrings, empty-results guard"

v0.1.21 — Phase 2 polish: .gitignore cleanup, docstrings, empty-results guard

v0.1.20 — Deep code review: 4 bug fixes, 31 tests, deps pinned, +1,312 images

v0.1.15 — Cache cleanup: 11 None entries purged across 6 cities

cache prewarm + concurrency improvements

v0.1.0 — Deterministic translation cache keys

v0.0.10 — Tokyo & Kyoto translation deep fix

v0.0.9 — Full cache sweep + adaptive radius fix

Skip media/entertainment Wikipedia pages when fetching attraction images

Persist translation cache to disk + prewarm script

Remove Tier 6 city-level photo fallback — show emoji instead of duplicate photos

Fix Wikipedia Commons thumbnail URLs: 400px -> 500px (valid size)

Improve image enrichment: city context in Tier 1 + smarter warmup health check

Unified warmup script + provider observability + Tokyo caches

v0.0.5 — Deterministic cache, Surprise Me, image fixes, health check

feat: pass extra_body={'think': False} for Ollama Cloud provider

feat: switch primary provider to Ollama Cloud DeepSeek V4 Flash

v0.0.3 — swap GPT-OSS 120B for Gemma 4 26B (free, 5x faster)

v0.0.2: multi-provider LLM rotation, city list trim, .env fix

v0.0.1