llm7: auto-truncate long prompts to 8000-char anonymous cap (no token); drop pro GLM model; keeps fast llm7 path for all prompt sizes

llm7: keep only confirmed-free models (mistral-small-3.2 lead), faster 45s timeout/failover

AutoFastLLM: free OpenAI-compatible providers (llm7.io pool + Pollinations/GAS fallbacks), lazy sticky routing, 429 cooldown, optional LLM7_API_KEY

Auto-lift prompt char cap when an llm7/HF proxy token is configured; document rate-limit fix (PROXY_KEY secret) in README

AutoFastLLM: throttle requests to respect proxy rate limit (1/s), exponential backoff on 429, and reject GAS deflection responses so garbage never surfaces as the answer

AutoFastLLM: shrink oversized prompts to fit the reliable gpt-4o-mini proxy instead of falling back to the search-style GAS proxy (fixes degenerate planning output)

Skip long prompts on limited HF proxy

Stabilize AutoFastLLM fallback routing

Fix Chatbot message format and tool alias

Add AgentFlow SearXNG AutoFast Space