Document the reasoning selector and single-session behaviour

README.md

@@ -18,9 +18,9 @@ Static chat client for the hosted OpenAI-compatible endpoint at `https://api.les
 - Model: `qwen3.8-27b-humanlike-chat`
 - No API key required
 - Optional system prompt for a persona, setting, or relationship, kept in the browser between reloads
-- New chat clears the conversation without reloading the page
-- Requests send only `model`, `messages`, `stream`, and `max_tokens`, so the endpoint defaults match the deployment gate
-- Thinking is off by default
+- Reasoning level selector: off (default), low, or medium. High and xhigh are rejected by the service
+- One conversation per page load; refreshing starts a new session
+- Requests send only `model`, `messages`, `stream`, `max_tokens`, and `chat_template_kwargs`, so the endpoint defaults match the deployment gate
 
 The service runs on scale-to-zero capacity. The first message after an idle period can take about two to three minutes while a worker starts, and the request stays connected while that happens. Later messages are fast until the worker shuts down after five idle minutes.