Component explorer
22 components across 12 kinds. Each entry documents: why it exists, how it works internally, head-to-head product comparison with pricing, when to use / not to use, failure modes, production usage with citations.
LLM providers
0/6 authoredOpenAI API (GPT-4o, GPT-4o-mini, o1, o3)
The reference LLM provider — pricing, capabilities, quirks.
Anthropic API (Claude 3.5 Sonnet, Opus, Haiku)
The frontier lab whose models most closely rival GPT-4 — and often beat it on reasoning.
Google Gemini API (2.5 Pro, Flash)
The multimodal-first LLM with the largest context windows in the industry.
Azure OpenAI Service
OpenAI models with enterprise compliance, regional deployment, and Microsoft billing.
AWS Bedrock
Multi-provider LLM gateway with IAM, VPC, and PrivateLink.
Llama, Mistral, Qwen, DeepSeek (open-weight models)
When self-hosting wins over API — cost, latency, control.
Embedding models
0/3 authoredOpenAI embeddings (text-embedding-3-small/large)
The default embedding model for RAG.
Cohere Embed (v3, multilingual)
The multilingual specialist and reranker vendor.
Voyage embeddings
The retrieval-optimized embedding vendor that often beats OpenAI on benchmarks.
Vector databases
0/5 authoredPinecone
The managed vector DB pioneer — pricing, features, ops story.
Weaviate
Open-source hybrid search with graph relations.
Qdrant
The Rust-based vector DB with strong filter performance.
pgvector (Postgres extension)
The 'just use Postgres' answer to vector search — and when it wins.
Chroma
The embedded / dev-first vector DB for prototypes.
MCP servers
0/1 authoredMCP servers (Anthropic Model Context Protocol)
The 2024 standard for LLM ↔ tool interoperability.
Agent frameworks
0/4 authoredLangChain
The most-adopted (and most-criticized) LLM framework — when to use, when to skip.
LlamaIndex
The RAG-focused alternative to LangChain.
LangGraph
Stateful multi-agent orchestration on top of LangChain.
CrewAI
Role-based multi-agent orchestration.
Eval tools
0/1 authoredRagas (RAG evaluation)
The default RAG-evaluation framework — faithfulness, answer relevance, context precision.
LLM observability
0/1 authoredLangfuse (LLM observability)
Open-source LLM tracing, prompt versioning, evaluation dashboards.
Guardrails
0/1 authoredNVIDIA NeMo Guardrails
The programmable-guardrails framework — Colang policies.