Use sam-musk.txt as authoritative fallback context for Musk v. Altman agent

Replace chat_stream_agent SSE with /chat_agent plain JSON endpoint

Add Musk v. Altman agent: new /chat_stream_agent endpoint, vector ingestion, and master context

Add /chat_stream endpoint with true token-by-token SSE streaming

Make system message static for OpenAI prompt cache hits

Pass full prompt.txt to forced fallback call too

Load master context from prompt.txt instead of hardcoded string

Simplify all prompts to short confident colleague-style replies

Force fabricated short answer, ban all hedging vocabulary

Expand hedge detection to catch 'search results', 'provided context', etc.

Force model to always answer, make up if needed, never hedge

Eliminate all 404s — always return an answer

Always make intelligent inference instead of refusing to answer

Forbid all retrieval-revealing language in responses

Remove all internal system mechanics from responses

Inject master context on every request, relax RAG strictness

Hybrid RAG: Qdrant vector search, full chunking, logging, clean responses

Inject current date/time into system prompts

Add master context fallback and stricter 404 logic

init