AI Parlance
v0.1 is specification-only. Parser, validator, and transpilers are not published yet. This documentation defines the language; code generation is planned.
.aip; transpilers turn the spec into concrete implementations.
Human or AI
↓
AI Parlance (.aip)
↓
AST + validation
↓
Transpilers
↓
Go, TypeScript, SQL, OpenAPI, …
Why it exists
When models generate code directly in general-purpose languages, they face:
- large amounts of boilerplate (CRUD, routes, validation, migrations)
- inconsistency across stacks
- large context → more tokens and more hallucination
- business rules and permissions scattered in generated code
AI Parlance concentrates intent in a compact semantic layer before implementation.
What it describes (and what it does not)
| Describes | Does not describe |
|---|
| entities, types, relations | custom UI |
| CRUD, indexes, migrations | complex algorithms |
| auth, policies | integrations without a dedicated block |
| domain workflows and events | hand-written imperative SQL |
Minimal example
app Demo @0.1 {
database postgres
}
entity User {
name: string required
email: email required unique
}
crud User
app block (Specification).
One definition can feed multiple generators (see transpiler matrix).
AI-first principles
The language prioritizes, in order:
- domain intent and semantics
- predictable structure for LLMs
- static validation before generation
- multi-target implementation via transpilers
Documentation map
| Page | Content |
|---|
| Specification | Grammar, builtins, policies, stability |
| Cost impact | Tokens, honest comparisons |
| Syntax | Core + Infra blocks |
| Database | Migrations, indexes, naming |
| Security | Auth, policies, rate limit |
| Workflows | Events, jobs, queues |
| Agents | Prompts, ai_context, best practices |
