mirror of
https://github.com/jacob-bd/openclaw-newsroom.git
synced 2026-08-14 00:47:57 +00:00
Major dedup overhaul — persistent SQLite database (dedup_db.py) stores normalized URLs and titles across all scans, replacing text-file-only matching that missed query-param variants and same-event rewrites. LLM editor now uses a 3-tier failover chain (Gemini Flash Lite → Grok via OpenRouter → Gemini Flash Preview) instead of single-provider with raw-fallback. Providers alternate to avoid double failure on outages. Inline AI keyword filter in the orchestrator blocks non-AI articles at the RSS extraction stage. Editorial rules updated from "must produce 5" to "up to 7, quality over quantity." Includes 68 unit tests, updated README, changelog, and migration guide.