Initial commit: Agentic OS troubleshooting platform

Self-hosted, Docker-based agentic troubleshooting platform: FastAPI backend +
LangGraph agent, Next.js UI, tiered LLM routing (local Ollama -> Gemini ->
DeepSeek -> OpenRouter), MCP server manager, encrypted device credentials,
RBAC, audit log, project-memory + Obsidian integrations, and editable
troubleshooting decision rules tuned for the GeneseasX vessel stack.

Co-authored-by: Cursor <cursoragent@cursor.com>

backend/app/services/llm.py

@@ -0,0 +1,24 @@
+"""LLM client builders — delegates tier selection to model_router."""
+from __future__ import annotations
+
+from app.config import settings
+from app.services.llm_usage import ModelInfo
+from app.services.model_catalog import entry_for_model
+from app.services.model_config import get_routing_config, resolve_tier_model
+from app.services.model_router import build_chat_model, entry_to_info
+
+
+async def triage_model():
+    cfg = await get_routing_config()
+    entry = resolve_tier_model(cfg, "local")
+    if not entry:
+        entry = entry_for_model("ollama", settings.llm_local_model, "local")
+    return build_chat_model(entry, temperature=0.0)
+
+
+async def triage_model_info() -> ModelInfo:
+    cfg = await get_routing_config()
+    entry = resolve_tier_model(cfg, "local") or entry_for_model(
+        "ollama", settings.llm_local_model, "local"
+    )
+    return entry_to_info(entry, "triage")