Skip to content

Technology Origins

This page tracks where the major Agentic Web Stack technologies came from, who originally developed or stewarded them, and why they matter to agent-native systems.

Dates are practical architecture dates: first public announcement, first stable specification, or major public release. Some open-source projects evolve continuously, so treat these entries as orientation, not legal or archival provenance.

How to Read This Page

The origin of a technology is not trivia. It helps estimate maturity, governance risk, interoperability expectations, and how quickly the interface may change.

  • A standards-body technology such as OAuth2, OpenID Connect, RDF, SPARQL, or RFC-based well-known URIs usually has slower change but broader institutional adoption.
  • A foundation-governed project such as A2A or OpenAPI usually signals an attempt to move beyond one company's implementation choices.
  • A company-originated API or framework can move quickly, but may need extra abstraction if vendor neutrality is important.
  • A university or research-originated runtime such as vLLM may be valuable because it encodes a specific systems idea, not only a product surface.
  • A community-maintained open-source tool such as pgvector or llama.cpp may become infrastructure because it is simple to adopt and easy to run locally.

Use this page with the Technology Stack: the stack page explains where a technology belongs, while this page explains where it came from and who currently shapes it.

Agent-Specific Protocols

Agent-specific protocols are the newest part of the stack. They should be evaluated with extra attention to versioning, governance, security model, SDK maturity, and implementation examples.

TechnologyFirst public milestoneOrigin / stewardWhy it matters
Agent2Agent (A2A)Announced by Google in 2025; donated to the Linux Foundation in June 2025; v1.0 announced as the first stable version in 2026Originally developed by Google; now a Linux Foundation project with a technical steering committeeProvides an agent-to-agent interoperability layer for discovery, task delegation, messages, artifacts, streaming, and asynchronous work
Agent CardEmerged with A2A discovery in 2025 and stabilized with the A2A specificationDefined by the A2A projectGives agents a machine-readable self-description at /.well-known/agent-card.json, including identity, skills, endpoints, capabilities, and security requirements
Model Context Protocol (MCP)Introduced by Anthropic in November 2024Originally developed by Anthropic; now broadly adopted across AI clients, tools, and serversStandardizes how AI applications and agents connect to tools, resources, and prompt templates

Web, API, and Identity Foundations

These are older web foundations. They are not agent-specific, but they are critical because agent systems still need ordinary service contracts, authentication, authorization, and metadata.

TechnologyFirst public milestoneOrigin / stewardWhy it matters
RESTDefined in Roy Fielding's 2000 doctoral dissertation at UC IrvineRoy T. Fielding, University of California, IrvineArchitectural basis for much of the modern web API ecosystem
OAuth2RFC 6749 published by the IETF in October 2012IETF OAuth Working GroupMain authorization framework for delegated, scoped access to protected resources
OpenID ConnectFinal OpenID Connect Core 1.0 specification published in 2014; errata set 2 incorporated in 2023OpenID FoundationAdds interoperable identity and ID tokens on top of OAuth2
OpenAPIOpenAPI Initiative launched in 2015 after SmartBear donated the Swagger Specification to the Linux FoundationSmartBear, OpenAPI Initiative, Linux FoundationMakes HTTP APIs machine-readable for humans, agents, validators, SDK generators, and gateways
GraphQLDeveloped at Facebook in 2012; open-sourced in 2015; moved under the GraphQL Foundation in 2018Facebook / Meta, GraphQL Foundation, Linux FoundationLets clients query precisely shaped data, useful when agents need flexible access to complex domains
JSON SchemaDraft 0 and Draft 1 published on 5 December 2009Early drafts by Kris Zyp and contributors; now community-maintained at json-schema.orgCommon schema vocabulary for validating and documenting JSON payloads

Memory, Retrieval, and Knowledge

Memory technologies come from two different traditions: semantic web and graph systems on one side, and embedding/vector retrieval systems on the other. Practical agent systems often combine both.

TechnologyFirst public milestoneOrigin / stewardWhy it matters
RDFW3C Recommendation in February 1999W3C, with early editors from Nokia Research Center and W3CFoundation for semantic web data and graph-based knowledge representation
SPARQLW3C Recommendation in January 2008; SPARQL 1.1 in 2013W3C SPARQL Working GroupQuery language for RDF graphs and semantic knowledge stores
Neo4jPrototype work began around 2000; first version in 2002; company and open-source release in 2007Neo4j founders Emil Eifrem, Johan Svensson, and Peter NeubauerMature native graph database for entity and relationship modeling
ChromaFounded in 2022; public open-source project and cloud service followedChroma, co-founded by Jeff Huber and Anton TroynikovDeveloper-friendly embedding and search infrastructure for AI applications
QdrantFounded in 2021André Zayarni and Andrey VasnetsovRust-based vector search engine and database for semantic retrieval
WeaviateLaunched by SeMI Technologies in 2019; company renamed Weaviate in 2023SeMI Technologies / WeaviateAI-native vector database with semantic and hybrid search features
pgvectorEarly 2020s open-source PostgreSQL extensionAndrew Kane and contributorsAdds vector similarity search to PostgreSQL, useful when teams want embeddings inside an existing relational database

Models, Local Runtimes, and Gateways

Model infrastructure changes faster than the web standards layer. For production systems, provenance helps distinguish hosted model APIs, local runtime projects, serving engines, and gateway abstractions.

TechnologyFirst public milestoneOrigin / stewardWhy it matters
OpenAI APIAPI announced in 2020; OpenAI founded in 2015OpenAIPopular hosted model API surface for text, multimodal, embeddings, tools, and agent workloads
ClaudeIntroduced by Anthropic in March 2023AnthropicHosted model family designed around helpfulness, reliability, and API access
GeminiGemini 1.0 announced by Google and Google DeepMind in December 2023Google / Google DeepMindHosted multimodal model family used across Google AI products and APIs
llama.cppReleased in March 2023Georgi Gerganov and the ggml communityMade local LLM inference practical on consumer hardware using C/C++ and quantized model formats
OllamaPublic project activity and Docker support appeared in 2023OllamaSimplifies running local models on macOS, Linux, and Windows
vLLMAnnounced by UC Berkeley researchers in 2023Sky Computing Lab at UC Berkeley; LMSYS ecosystemHigh-throughput LLM serving engine built around PagedAttention
LiteLLMPublic open-source AI gateway in the early LLM provider-fragmentation eraBerriAI / LiteLLM contributorsProvides a unified OpenAI-compatible interface, proxy, routing, logging, and budget control across many model providers

Agent Frameworks

Agent frameworks sit closest to application code. They are useful for orchestration and developer productivity, but they should not replace protocol-level interoperability contracts when systems need to cross vendor or organizational boundaries.

TechnologyFirst public milestoneOrigin / stewardWhy it matters
LangGraphLaunched by LangChain in early 2024LangChainStateful, graph-based orchestration for long-running and controllable agents
CrewAIEmerged as a public multi-agent framework around 2023-2024Created by João Moura; CrewAI company and communityRole-based multi-agent workflows, crews, flows, memory, and observability
Semantic KernelPublic Microsoft documentation and SDK work became visible in 2023; agent framework work matured laterMicrosoftModel-agnostic SDK and agent framework for .NET, Python, and Java ecosystems
Agent Development Kit (ADK)Introduced by Google at Cloud Next 2025GoogleOpen-source framework for building, debugging, evaluating, and deploying agents and multi-agent systems

References

Page created by Dr. C. Klukas