Describe what you want. Get a live app at a URL. Share it with the world. AI-generated software as a social platform.
Before YouTube, sharing video meant hosting files, managing bandwidth, picking codecs. YouTube made it: record, upload, share a link.
AI agents generate enormous volumes of small, purpose-built software. But every piece still requires the creator to think about deployment, scaling, security, and infrastructure.
Compute, storage, databases — all commodity and swappable. What doesn't exist is the software layer: the experience, the network effects, the social composability.
That's the equivalent of asking every YouTube creator to run their own CDN. The infrastructure is commodity. The platform layer is the product.
The fundamental experience — the "YouTube upload moment" for software:
"I need a budget tracker." The creator describes what they want in natural language. No code, no config.
The platform calls the LLM (OpenAI gpt-4o), generates HTML/JS/CSS, and extracts the result. A mock generator works offline for development.
Files are written to data/apps/{subdomain}/. CSP headers and security policies are injected automatically. The creator never touches infrastructure.
The app is immediately accessible at {name}.softcloud.localhost. No signup required to use it — consumption is just a URL.
Every app lives in a Space (like a YouTube channel). Sharing and forking are built into the data model from day one — not bolted on later.
A Go backend compiled from standard library packages — no external dependencies. Designed for swappability at every layer.
:8080 — REST endpoints for creating, listing, and retrieving apps
:8081 — Subdomain routing, static file serving, CSP injection
OpenAI gpt-4o or mock — swappable via interface
Data types: App, Space, CSPPolicy, Manifest. Pure structs, no logic beyond CSP rendering.
Persistence interface + in-memory implementation. Thread-safe with sync.RWMutex.
Generator interface with OpenAI and Mock implementations. Extracts HTML from markdown code fences.
Deployer interface + filesystem implementation. Writes files to disk, returns absolute paths.
Subdomain routing via Host header. Injects CSP, X-Frame-Options, X-Content-Type-Options on every response.
HTTP server, router, CORS middleware. Wires store + generator + deployer into REST handlers.
SoftCloud is designed to avoid the Vercel trap (runtime first, social later) and the Glitch trap (social without great generation).
Every app lives in a Space — the creator's workspace, like a YouTube channel. Spaces are structural, not UI chrome. The data model includes SpaceID on every app from birth.
The platform knows an app's lineage — was it forked from another app? What agent session created it? This enables a remix culture where software builds on software.
Using an app requires nothing: no signup, no install, no SDK. Just a URL. Authentication only when the app itself demands it.
Unlike media, software composes. The platform will provide standard idioms for apps to communicate and share data — a composition layer where apps build on apps.
"Worse is better" — Optimize for simplicity and reach over power and control. A constrained but frictionless experience beats a powerful but complex one. RealPlayer had more features; YouTube won.
Creators don't think about security any more than YouTube creators think about CDN topology. SoftCloud enforces it structurally.
Each app runs at its own subdomain — a separate browser origin. localStorage, cookies, and DOM are isolated by the browser itself. Cross-origin fetch blocked by CORS.
script-src 'self' — no inline scripts, no eval(). frame-src 'none' — no iframes. object-src 'none' — no plugins. Injected on every response.
A documented contract specifying what generated apps are allowed to do, what CDNs are whitelisted, and what the security headers enforce. Gaps are documented honestly.
SoftCloud builds the six layers that sit between generated software and commodity infrastructure:
What users see: intent input, app URLs, spaces, sharing, discovery
Intent-to-code generation, lifecycle management, evolution, composition
Identity, storage, analytics, payments, inter-app plumbing
Moderation, security, compliance, resource limits, runtime monitoring
Unified interface to backend services — makes infrastructure swappable
Commodity compute, databases, storage, CDN — swappable
Repository: /home/samschillace/dev/ANext/softcloud
git log --oneline | wc -lSTATE.yaml (completed_features list)find . -name '*.go' | xargs wc -l (excluding test files for source, test files only for tests)STATE.yaml meta sectiongit shortlog -sne --allgit log --format="%ai" (first and last)wc -l specs/**/*.mdKey source documents:
README.md — project summaryPROJECT-INTENT.md — vision, user stories, definition of donedocs/design/2026-04-20-soft-cloud-design.md — original design document (358 lines)specs/architecture.md — architecture specification (686 lines)STATE.yaml — project state, feature tracking, blockersFEATURE-ARCHIVE.yaml — completed feature timestampsbackend/cmd/softcloud/main.go — composition root (60 lines)All quotes are from project source documents. No metrics are fabricated. This is an early-stage project — the "Not Yet Built" list is longer than the "Built" list by design. Phase 1 is about proving the core loop.
Primary contributor: samschillace · Category: Showcase · Generated: May 2026