nakul-bhai/group-chat

Switchable primary provider via LLM_PRIMARY_PROVIDER env var

Change Groq model to openai/gpt-oss-20b

Add gitignore, remove cached pycache files

Add Groq as fallback LLM provider for timeout resilience

Fix reply ordering and add debug logging

Fix: Allow cross-mention to trigger bots who already replied

Fix: Show speaker names in history and previous replies

Restore detailed XML prompts + add group context to prevent loops

Fix: Make prompts natural and casual, remove forced questions

Fix: Reduce per-agent timeout to 8s for faster response

Fix: Run primary first, then others in parallel (faster response)

Fix: Sequential bot execution so each sees all previous replies

Feat: Random agent selection + cross-bot @mention triggering

Fix: Always loop replies back to user to prevent bot-only echo

Fix: Sequential agent execution with @mention group chat awareness

Fix: Remove sample dialogues, raise temperature to 0.95, increase tokens to 250

Fix: Move SSE pushes to background, add SSE connected event

Fix: Return all agent replies with emotions & psychology-enhanced prompts

Use query param ?character=X for per-character routing

Add per-character endpoints for individual webhooks

Fix: gather all replies, push SSE directly, use correct service name

Strip HTML details tags from ALL LLM responses

Simplify agent logic: always reply with role context

Force 3 replies per message with jealousy dynamics

Fix timeout: return first reply immediately, collect rest in background

Defensive sort: filter replies missing required keys

Fix force_reply return missing priority and service keys

Update Tomoko prompt with roleplay action format

Update Hyouka prompt with roleplay action format

Update Chika prompt with roleplay action format

Update Hakuo prompt with roleplay action format

Fix JSON parser to handle Qwen HTML details tags

Add HF Space metadata to README

Initial commit: Crazy Ones Group Chat

initial commit