Always-on grounded retrieval with SearxNG and role-scoped Ollama models (#62).

Phases 1–3: split THINKING/FAST/UTILITY/EMBED models, structured search with
SearxNG primary + DDGS failover, fail-open grounding, citations on Prompt + WS
frames, single history window with prompt budgeting, and reindex_embeddings.

.env.example

@@ -12,8 +12,22 @@ DATABASE_URL=postgres://chat_backend:chat_backend@db:5432/chat_backend
 
 # Ollama — local loopback when Ollama runs on this machine; LAN IP for GPU host.
 OLLAMA_BASE_URL=http://127.0.0.1:11434
-# OLLAMA_MODEL=llama3.2
-# OLLAMA_EMBED_MODEL=llama3.2
+# Legacy fallback (used when role-specific vars unset). Prefer the role vars.
+# OLLAMA_MODEL=gpt-oss:20b
+# OLLAMA_MODEL_THINKING=gpt-oss:20b
+# OLLAMA_MODEL_FAST=gemma4:latest
+# OLLAMA_MODEL_UTILITY=llama3.2
+# OLLAMA_EMBED_MODEL=nomic-embed-text
+# OLLAMA_NUM_CTX_THINKING=16384
+# OLLAMA_NUM_CTX_FAST=8192
+# OLLAMA_NUM_CTX_UTILITY=4096
+
+# Web search (#62) — SearxNG primary, DDGS failover. See README "SearxNG".
+ALLOW_INTERNET_ACCESS=true
+SEARCH_PROVIDER=searxng
+SEARCH_FAILOVER_PROVIDER=ddgs
+SEARXNG_BASE_URL=http://127.0.0.1:8080
+# SEARXNG_TIMEOUT_SECONDS=8
 
 # Email (SMTP2GO) — optional for local
 EMAIL_HOST=mail.smtp2go.com