Orieon · Code-Backed Runtime & Compiled Target Architecture Diagram

AI chat path found in source

SearchPanel classifies each question, runs keyword plus semantic retrieval, and requests a grounded answer when needed.
Rust keeps one Qwen worker resident; qwen_worker_tool serves localhost /classify and /rag and caches the active per-video index.
If the coordinator fails, the UI falls back through orieon.exe to local_classify_tool.exe and local_rag_tool.exe.

CMake compiled target map

orieon.exe links local_rag_core, local_rag_llm, llama.cpp, ONNX Runtime, SentencePiece, DXGI and WinHTTP.
local_rag_core = types + index + chunk store + embedder + retriever; local_rag_llm = LocalLlm + GPU hardware detection.
Shipped helpers include index, advanced search, RAG, summary, classify, resident Qwen worker and resident session helper; test-only targets are separate.

Cloud correction from Python code

The backend is FastAPI and is written for Cloud Run-style deployment, with Neon PostgreSQL, Upstash Redis and Cloudflare R2 integrations.
Identity Platform, Firestore/Spanner, GCS and Cloud Armor are not implemented in the inspected source and are intentionally omitted.
The source copy has incomplete .git metadata, so this diagram is working-tree audited but cannot provide commit-pinned evidence links.