Hourly source-led research portal. Each dispatch starts from one of 24 topic families, finds a new current angle, gathers evidence through SearXNG, and publishes a linked synthesis.
- Schedule: hourly (
0 * * * *) - Local model: discovered dynamically from
/v1/modelsat192.168.0.219:8080/v1 - Local inference: DeepSeek V4 uses native thinking-disabled streaming; other models retain the legacy payload; each response waits up to 20 minutes
- Local status:
/home/pi/.hermes/cron/markgitup-local-inference-status.json - Fallback LLM: GPT 5.6 Luna through Hermes
openai-codex - Failure policy: no post is written when both inference paths fail
- Search: local SearXNG at
127.0.0.1:8888 - Source gate: publish only after at least 2 sources; low-source searches select another unused topic family (up to 6 attempts)
- Output: GitHub Pages via
main