feat(ai-chat): use openai/gpt-oss-120b and token-budget the role context

Switch the assistant model from qwen/qwen3-32b to openai/gpt-oss-120b. Cap every per-role context list, hard-clamp the assembled system prompt to ~4000 tokens at a line boundary, and bound re-sent chat history to 12 messages so the model context window can't be exceeded regardless of mock-data volume.
2026-05-30 20:33:01 +05:30
parent 034ed1d180
commit 2ce3395006
2 changed files with 57 additions and 15 deletions
@@ -99,7 +99,7 @@
 - **Maps**: react-leaflet
 - **Charts**: Recharts
 - **Icons**: lucide-react
- **AI**: Groq SDK (qwen-32b) with dynamic role-aware system prompts
+- **AI**: Groq SDK (openai/gpt-oss-120b) with dynamic role-aware system prompts (token-budgeted)
 - **Notifications**: sonner (toasts)

 ### Setup