[Sync] Lua Hyprland chain, licence cleanup

wm/.config/aichat/config.yaml

@@ -1,5 +1,6 @@
-# WM HOST OVERLAY — full-file copy of common/.config/aichat/config.yaml with the default model
-# flipped to the duskadiy client: wm is not on the home LAN, so the `local` endpoint
+# WM HOST OVERLAY — full-file copy of common/.config/aichat/config.yaml plus the `wm` client
+# (this box's own llama.cpp router at 127.0.0.1:11343, see wm/.config/llamacpp/README.md) and the
+# default model flipped to the duskadiy client: wm is not on the home LAN, so the `local` endpoint
 # (192.168.0.204) is unreachable here (kept below for VPN use). Stowed with --override so it
 # shadows the common file on wm only. When the common config changes, mirror the change here.
 # see https://github.com/sigoden/aichat/blob/main/config.example.yaml
@@ -21,6 +22,34 @@ right_prompt: '{color.purple}{?session {consume_tokens}/{max_input_tokens} }{col
 
 # max_input_tokens = real ctx-size (from the router) minus output headroom.
 clients:
+# This box's own GTX 1080 Ti server (`llamaserver` abbr → 127.0.0.1:11343; presets + benched
+# speeds in wm/.config/llamacpp/README.md). Model id = config.ini section name.
+- type: openai-compatible
+  name: wm
+  api_base: http://127.0.0.1:11343/v1
+  models:
+  - name: Qwen3.5-9B-Q4_K_M
+    max_input_tokens: 14000            # ctx 16384 · 33 t/s — daily driver, instant answers
+    supports_vision: true
+  - name: Qwen3.5-9B-Thinking
+    max_input_tokens: 14000            # ctx 16384 · same file, reasoning on (budget 8k)
+    supports_vision: true
+  - name: Qwen3.5-9B-MTP
+    max_input_tokens: 7000             # ctx 8192 · 57 t/s (MTP, +0.75 GB VRAM) — loads on a QUIET desktop only
+    supports_vision: true
+  - name: gemma-4-E4B-it-UD-Q4_K_XL
+    max_input_tokens: 30000            # ctx 32768 · 85 t/s (MTP) — fastest; small vision model, long docs
+    supports_vision: true
+  - name: Mellum2.1-12B-A2.5B-Thinking-IQ4_XS
+    max_input_tokens: 14000            # ctx 16384 · 63 t/s (87 quiet) — fast reasoning coder, prefill > 1k t/s
+  - name: Ornith-1.5-35B-A3B-IQ4_XS
+    max_input_tokens: 14000            # ctx 16384 · ~10 t/s (MTP, 47% acc) — agentic coder, thinking
+  - name: Ornith-1.5-35B-A3B-instant
+    max_input_tokens: 14000            # ctx 16384 · 14–21 t/s busy / 31 quiet (MTP) — same file, thinking off: agent loops
+  - name: gpt-oss-20b
+    max_input_tokens: 30000            # ctx 32768 · 17 t/s — reasoning + tools (TTFT = thinking)
+  - name: gpt-oss-20b-low
+    max_input_tokens: 30000            # ctx 32768 · reasoning effort low, snappy
 # LAN access (fast; only reachable on the home network)
 - type: openai-compatible
   name: local
@@ -54,7 +83,7 @@ clients:
   - name: GLM-4.7-Flash-UD-Q3_K_XL
     max_input_tokens: 22000            # ctx 24576 · ~28 t/s — fast coder (3-bit)
   - name: Ornith-1.5-35B-A3B-IQ3_XXS
-    max_input_tokens: 22000            # ctx 24576 · ~32 t/s — agentic coder, thinking
+    max_input_tokens: 30000            # ctx 32768 · ~32 t/s — agentic coder, thinking
     supports_vision: true
   - name: gpt-oss-20b
     max_input_tokens: 62000            # ctx 65536 · ~38 t/s — fast reasoning + tools