Add Gemini model fallback cascade for 429 rate limits - #15

Merged

allocsys merged 6 commits into

mainfrom

feature/gemini-connector

Jul 25, 2026

Commits on Jul 25, 2026

Fix gemini_investigate: use role 'user' for functionResponse turns (Gemini API rejects role 'function')
allocsys
committed
8c261ea
Add notion_get_page to gemini_investigate's delegated function set, so investigations can actually read page content (previously only notion_search/notion_query_database were available)
allocsys
committed
8427eb8
Raise gemini_investigate HARD_MAX_STEPS from 10 to 20 -- Vercel Fluid Compute defaults to a 300s function timeout, giving far more headroom than the original 10-step cap assumed
allocsys
committed
b1658a8
Update gemini_investigate tool description: hard cap is now 20, not 10
allocsys
committed
8d4866c
Add GEMINI_FALLBACK_MODELS config for rate-limit cascade
allocsys
committed
7045b23
Cascade through GEMINI_FALLBACK_MODELS on 429 rate-limit errors, since free-tier quotas are tracked per model
allocsys
committed
f6fe5fa