Most recent
navigate open esc close Corpus index built 2026-06-07 23:58 UTC

§ TOPICS

Topics

A production LLM deployment is six layers of infrastructure, not one model. Pick a layer below, then a category inside it, to see every survey, case, and disclosure NuClide has on that platform class. 446 artifacts across 37 categories.

§ Reference topology

An AI/LLM application is nine layers deep.

Drawn as a layer cake from the user down to the public internet. Each layer has its canonical implementations and a per-layer population count from the corpus. Magenta-bordered nodes ship insecure-by-default on at least one popular distribution.

Chat UIs layer 09 · user
Open WebUI AnythingLLM LobeChat LibreChat custom front-ends
3,400+ unauthenticated chat front-ends
Agent / RAG APIs layer 08 · orchestration
LiteLLM LangServe LangFlow Flowise custom routers
1,200+ open Agent / RAG endpoints
Model servers layer 07 · inference
Ollama llama.cpp vLLM TGI Triton LocalAI
16,473 unauthenticated Ollama · 1,200+ vLLM
Vector DBs layer 06 · retrieval
Qdrant Milvus Weaviate Chroma Pinecone (hosted)
2,100+ open vector indices
Search / docs layer 05 · retrieval
Elasticsearch ClickHouse Solr Meilisearch Typesense
5,037 ES with dense_vector schema
Browser automation layer 04 · agents
Browserless Selenium Grid Playwright CDP proxies ComfyUI
548 unauthenticated ComfyUI · 6 live CDP sessions
Data layer layer 03 · storage
Postgres MongoDB MinIO / S3 Redis etcd Vault
3,014 etcd · 912 Vault · 4,105 Consul
Orchestration layer 02 · compute
Kubernetes Docker Compose Nomad systemd
Docker defaults are the proximate cause across most layers above
GPU compute layer 01 · hardware
H100 H200 L40S A100 RTX 5090 consumer cards
10× L40S in one fleet observed
layer 00 · the public IPv4 internet