Primitives.org.ai
The fundamental abstractions for building AI-native software
Build Business-as-Code & AI-Delivered Services-as-Software
Primitives for building software where humans and AI work together seamlessly. From functions to businesses, every one of them is designed for the age of AI.
Function
Code, Generative, Agentic, Human—unified interface
Database
Schema-driven with vector embeddings
Workflow
Durable execution and state machines
Task
Work units with assignment and tracking
Agent
Autonomous entities with real identity
Human
Approvals, reviews, and judgment calls
Product
Compose primitives into products
Service
AI delivering human-quality work
Business
Organizations as executable code
How it is written down
Two of the primitives are not runtime objects but the format everything else is authored in, and they are documented on their own sites rather than restated here:
- MDXLD — our extension of MDX. MDX is an
open standard authored by the MDX community, not by us; MDXLD is ours, a superset
of MDX whose frontmatter is in turn a superset of YAML-LD. Every valid MDX document is
already a valid MDXLD document. The
$idand$typekeys are not our invention either — they come from the linked-data community's dollar-convenience context. Our one real deviation is$context, which is alwayshttps://schema.org.ai, unversioned. Also served at mdxld.org. - MDXUI — components and layouts selected from a document's
$type. Its home (ui.org.ai) is not yet serving; until it does, the packages are the documentation.
Built for the AI Era
The age of AI requires new primitives. Traditional software abstractions assume deterministic execution and human-only workflows.
- Heterogeneous Computation — Code, AI models, autonomous agents, and humans all participate in the same workflows with the same interfaces.
- Uncertain Execution — Generative and agentic functions produce variable outputs. The primitives handle retries, validation, and fallbacks transparently.
- Distributed State — Context flows across systems, conversations, and time horizons. Vector embeddings enable semantic understanding.
- Collaborative Work — AI and humans hand off tasks seamlessly. Each contributes their strengths with clear escalation paths.
Get Started
npm install ai-functionsStart with Functions—the foundation everything else builds on.