server: OpenAI API and llama.cpp Web UI compatibility

- OpenAI-style /v1/models, /v1/chat/completions (SSE), error bodies, CORS
- GET /props and enriched model rows for ggml Web UI
- Session file persistence and optional TILES_BOOTSTRAP_* startup load
- POST /models/load when TILES_MODEL_CACHE_PATH is set (Web UI load)
- Chat SSE: timings (Web UI) and metrics (Tiles CLI memory mode)
- Pytest coverage; TILES_SKIP_SESSION_PERSIST in tests
- just webui-llamacpp via scripts/phase2_llamacpp_webui.sh

Made-with: Cursor

Author: feynon

justfile

@@ -15,6 +15,14 @@ serve:
     server/.venv/bin/python3 -m server.main
     # uv run --project server python -m server.main
 
+# Python server (OpenAI compat API tests; requires uv sync in server/)
+test-server:
+    uv run --project server pytest server/tests/ -v
+
+# llama.cpp SvelteKit UI (clone under ../llama.cpp by default); proxies to Tiles :6969
+webui-llamacpp:
+    bash scripts/phase2_llamacpp_webui.sh
+
 bundle:
     ./scripts/bundler.sh
 

scripts/phase2_llamacpp_webui.sh

@@ -0,0 +1,43 @@
+#!/usr/bin/env bash
+# Run ggml-org/llama.cpp SvelteKit web UI with the dev proxy pointed at the Tiles server.
+#
+# Prerequisites: Node.js (npm), git. Start Tiles separately (e.g. just serve on :6969).
+#
+# Environment:
+#   LLAMA_CPP_ROOT   Clone path (default: sibling of this repo: ../llama.cpp)
+#   TILES_BACKEND    Proxy target (default: http://127.0.0.1:6969)
+
+set -euo pipefail
+
+ROOT="$(cd "$(dirname "$0")/.." && pwd)"
+LLAMA_CPP="${LLAMA_CPP_ROOT:-$ROOT/../llama.cpp}"
+BACKEND="${TILES_BACKEND:-http://127.0.0.1:6969}"
+WEBUI="$LLAMA_CPP/tools/server/webui"
+VITE_CFG="$WEBUI/vite.config.ts"
+
+if [ ! -d "$LLAMA_CPP" ]; then
+  echo "Cloning llama.cpp -> $LLAMA_CPP"
+  git clone --depth 1 https://github.com/ggml-org/llama.cpp "$LLAMA_CPP"
+fi
+
+if [ ! -f "$VITE_CFG" ]; then
+  echo "Expected file missing: $VITE_CFG"
+  exit 1
+fi
+
+if ! grep -qF "$BACKEND" "$VITE_CFG" 2>/dev/null; then
+  echo "Patching $VITE_CFG to proxy API routes to $BACKEND"
+  cp "$VITE_CFG" "${VITE_CFG}.tiles.bak"
+  perl -pi -e "s|http://localhost:8080|${BACKEND}|g" "$VITE_CFG"
+fi
+
+echo "Backend proxy: $BACKEND (run Tiles with just serve). Load model via POST /start, session file, TILES_BOOTSTRAP_*, or TILES_MODEL_CACHE_PATH + UI load."
+echo ""
+
+cd "$WEBUI"
+if [ -f package-lock.json ]; then
+  npm ci
+else
+  npm install
+fi
+exec npm run dev