WisdomHub, technical docs

Install (fresh VPS)

1. Copy the release zip to /var/www/wisdomhub and unzip.
2. Put your source PDF at data/book.pdf.
3. Edit config.local.php: set oneminai_api_key. The admin_token is auto-generated.
4. ./run.sh ingest, wait for it to finish (single-digit minutes without pre-warm, up to 20 with).
5. systemctl enable --now wisdomhub.
6. Include nginx.wisdomhub.conf in your existing sevinhub server block, nginx -t && systemctl reload nginx.
7. Open https://wisdom.sevinhub.com/.

Ingest flags

• ./run.sh ingest --skip-vision, skip image captioning (faster, no vision cost).
• ./run.sh ingest --skip-prewarm, skip the cache warming pass.
• ./run.sh ingest --chat-model gpt-4o-mini, cheaper model for pre-warm.

Endpoints (behind /wisdomhub/api/proxy.php)

• ?path=ask POST {question}, returns {answer, cited_pages, source, cached}. Source is exact, fuzzy, or live.
• ?path=tts POST {text, voice?}, streams cached MP3.
• ?path=health GET, returns chunk + cache counts.

Admin stats

curl -H "X-Admin-Token: $TOKEN" http://127.0.0.1:8088/stats shows chunk count, cache hits, and 1min credits spent. Get $TOKEN from config.local.php.

Rate limits

Defaults: 20 questions per IP per hour, 100 per day. Change rate_limit_hour and rate_limit_day in config.local.php.

Update

Drop the new release zip on top of the old install (keeping your config.local.php and data/ directory). Run ./run.sh ingest only if the PDF or embeddings changed.

Reset the cache

./run.sh ingest wipes chunks, chunks_vec, qa_cache, and qa_vec before rebuilding. Rate-limit history and usage log survive.