mirror of
https://github.com/hanzoai/chat.git
synced 2026-09-30 20:51:17 +00:00
zooqueen
47ad491722
The guest model was a paid house SKU (zen5-flash) that draws on the deployment balance, so with the balance overdrawn a guest's first message returned 402 and signed-out chat did not work at all. Pin the guest to a free route (nvidia/nemotron-nano-9b-v2:free) — the same free id @hanzo/ai names for a signed-in fallback — so a guest is served at zero cost regardless of balance. Guest scope now overrides the endpoint/model to the free one rather than rejecting a mismatch, so a stale client model can never leave a guest unable to send, while a guest still can never reach a paid route. The free guest model and endpoint are published in startup config so the client can open and name the conversation on it. Bumps @hanzo/ai to 0.3.4.