AI Universe · Place the new thing

Place the new thing

A new name appears. Before reading its docs, ask four questions. Most of the time you discover you already understand it: it is a known concept, done a new way.

1 · Which family?where does it sit in the map2 · Which bucket?what job does it do3 · What kind?implementation · bundle · protocol· wrapper · model · pattern4 · What do I already know?the concept's rules apply to it
Ask these four about any new name. Most of the time the answer to 4 is: a lot.

32 names, already placed

NameFamilyBucketKindYou already know it as
OllamaCompute & infrastructureServing engineWrapperan easy way to run open models locally (over llama.cpp)
vLLMCompute & infrastructureServing engineImplementationsoftware that serves a model fast
PineconeDataKnowledge and retrievalImplementationa hosted vector database
RAGDataKnowledge and retrievalPatternnot a product: the pattern of retrieving passages into the context
pgvectorDataKnowledge and retrievalImplementationa vector store inside PostgreSQL
DatabricksDataLakehouse layersBundle / platforma platform spanning data, models, serving, agents and governance
LoRAModelFine-tuning and alignmentImplementationfine-tuning by training a small add-on, not the whole model
Llama 3 / gpt-ossModelThe modelsModelopen-weight models you can host yourself
Azure API ManagementModel accessGatewayImplementationa general API gateway with AI policies
LiteLLMModel accessGatewayWrapperone OpenAI-shaped door over many providers' APIs
GroqModel accessInference APIBundle / platformcustom chips plus an inference API, sold as a service
Microsoft FoundryModel accessInference APIBundle / platformAzure's platform of models, agents, evaluation and safety
OpenRouterModel accessRouting and model choiceBundle / platformone key, many models, with routing
AWS Bedrock AgentCoreThe AI application (agent harness)Control loopBundle / platformAWS's set of harness parts: runtime, memory, identity, gateway, observability
Claude Agent SDKThe AI application (agent harness)Control loopBundle / platforma loop, tools, context and permissions, packaged by one vendor
Google ADKThe AI application (agent harness)Control loopImplementationGoogle's agent loop library
LangGraphThe AI application (agent harness)Control loopImplementationa control loop plus state, written as a graph
Microsoft Foundry Agent ServiceThe AI application (agent harness)Control loopBundle / platforma hosted harness: loop, tools, memory, tracing on Azure
Pydantic AIThe AI application (agent harness)Control loopImplementationa typed loop and a typed model layer
DBOSThe AI application (agent harness)Durable work and peopleImplementationdurable workflows that wait for a person
TemporalThe AI application (agent harness)Durable work and peopleImplementationthe same concept as DBOS, with its own server
Azure AI Content SafetyThe AI application (agent harness)Guardrails and policyImplementationharm and jailbreak checks, as an Azure service
Bedrock GuardrailsThe AI application (agent harness)Guardrails and policyImplementationinput and output checks, as an AWS service
A2AThe AI application (agent harness)Tools and protocolsProtocol / standardthe same idea as MCP, but agent to agent
MCPThe AI application (agent harness)Tools and protocolsProtocol / standardan agreed plug shape between an agent and its tools
Claude CodeExperienceCoding agentsBundle / platforma coding agent: an experience built on a full harness
Databricks GenieExperienceConversational analyticsImplementationask your data in words, over a semantic layer
DeepEvalAssurance & operationsEvaluationImplementationa library of graders and test cases
LangSmithAssurance & operationsEvaluationBundle / platformtracing, datasets and evaluation in one hosted product
LangfuseAssurance & operationsObservabilityImplementationtraces of every step, plus scores (it also covers evaluation)
OpenTelemetryAssurance & operationsObservabilityProtocol / standardthe standard way any code reports spans and metrics
PyRITAssurance & operationsSecurity and red-teamingImplementationan automated attacker for your agent

Some names sit in several buckets: a bundle like Microsoft Foundry or Databricks spans many families. Place it by the job you are using it for.

Compute & infrastructureDataModelModel accessThe AI application (agent harness)ExperienceAssurance & operations