Add grounded Ollama rewrite flow for chatbot

README.md

@@ -57,6 +57,50 @@ Admin auth:
 
 The app will be available at `http://localhost:4173`.
 
+## Optional local LLM (Ollama) for grounded rewrites
+
+You can keep deterministic retrieval as the source-of-truth and optionally rewrite responses with a local model.
+
+1. Install and run Ollama on your host.
+2. Pull a small model suited to older hardware, for example:
+
+```bash
+ollama pull qwen2.5:3b-instruct
+```
+
+3. Start the API with these environment variables:
+
+```bash
+CHATBOT_LLM_ENABLED=true
+CHATBOT_LLM_BASE_URL=http://127.0.0.1:11434
+CHATBOT_LLM_MODEL=qwen2.5:3b-instruct
+CHATBOT_LLM_TIMEOUT_MS=25000
+CHATBOT_LLM_NUM_CTX=2048
+```
+
+4. Call the rewrite endpoint from your existing chat flow:
+
+`POST /api/chatbot-grounded-rewrite`
+
+Request payload shape:
+
+```json
+{
+	"question": "Who was Titus?",
+	"draftAnswer": "Deterministic answer produced by current retrieval/synthesis.",
+	"sources": ["Episode 2 - Introduction to Titus"],
+	"contextChunks": [
+		{
+			"title": "Episode 2 - Introduction to Titus",
+			"sourceLabel": "Episode 2",
+			"content": "Titus was a Gentile..."
+		}
+	]
+}
+```
+
+If the endpoint fails or is disabled, keep your deterministic answer and existing fallback behavior.
+
 Persistent admin saves:
 
 - Admin updates are written to `data/admin-content.json`.