Structured Intent
Define what you are building, not just how. SODL captures the why so AI understands context, purpose, and scope — before writing a single line of code.
Active Development
Turn AI code generation from guesswork into engineering
SODL gives AI agents structured intent, constraints, and architectural decisions — so they generate the right code, every time.
Everything you need for structured AI generation
SODL gives you the primitives to express what you're building with enough precision for AI to get it right — every time.
Enforceable Policies
Group rules under named policies with severity levels (critical, high, medium, low). The AI must respect every constraint — not just the ones you remember to repeat.
Generation Pipeline
Orchestrate multi-step code generation with named steps, gates, and validation. Each stage has clear outputs and pass criteria before the next begins.
Framework Agnostic
Works with any tech stack. TypeScript, Python, Rust, Astro, Django — you declare the stack and let the AI fill in the rest.
Reusable Templates
Define base templates for common architectures and extend them across projects. Share stack choices and policy rules across your whole team.
AI-Native Design
Built from the ground up for Cursor, Claude, GPT-4, and the next generation of AI coding agents. Structured context, not free-form prompts.
How SODL works
A five-stage workflow that transforms vague intent into validated, precise AI generation.
Intent
Define what you are building and why — the primary goal, outcomes, and what is explicitly out of scope.
Architecture
Specify the structure: stack, interfaces, modules, and how components interact.
Constraints
Enforce policies and invariants with severity levels — rules the AI must never violate.
Compile
Run sodl compile to validate the spec — dependency graph, interface contracts, and circular dependency checks.
Generation
Feed the spec to Cursor, Claude, or GPT. AI produces the right code every time — no guesswork.
Built for AI-first developers
Whether you use Cursor, Claude, or GPT — SODL gives your AI agent the structure it needs to generate correct code.
- Cursor developers Use SODL specs as Cursor rules context for precise, consistent code generation across your entire project.
- Claude users Give Claude structured architectural intent instead of free-form prompts. Get reproducible, constraint-aware results.
- GPT builders Chain GPT-4 calls through a validated pipeline with invariant checking and gate conditions — no more prompt guesswork.
SODL in action
One spec file captures everything an AI needs to build your project correctly.
my-site.sodl
# Reusable base for Astro/Tailwind sites
template "AstroSiteBase":
stack:
language = "TypeScript"
styling = "Tailwind CSS 4"
hosting = "Vercel"
policy Accessibility:
rule "All interactive elements have aria attributes" severity=high
# Marketing + docs website
system "MySite" extends "AstroSiteBase":
version = "1.0.0"
stack:
framework = "Astro 5"
documentation = "Starlight 0.32"
intent:
primary = "Launch page for my SaaS product"
outcomes = [
"Hero with value proposition and CTAs",
"Feature grid with 6 key benefits",
"Pricing section",
"Blog powered by Starlight"
]
out_of_scope = ["User auth", "Payment processing"]
module UIComponents:
implements = [HeroComponent, FeatureGridComponent, PricingComponent]
artifacts = ["src/components/"]
pipeline "Production":
step Build:
require = "astro build"
gate = "Static site built to dist/"
step Deploy:
require = "Push to main triggers Vercel"
gate = "Site live with valid OG metadata"Browse the docs
Everything you need to go from first spec to production-ready AI generation.