RUNTIME
- Runtime Bun
- Containers Podman (rootless)
- Deployment Container / Serverless
born to be fast & light
Velociraptor
Instantiate Through Emulation
ETYMOLOGY
The letters v and r bracket 10 letters in Velociraptor, forming the abbreviation v10r.
Containerized Full-Stack Pattern Library
Ocean · Rounded
CAPABILITY TAXONOMY
A combinable technology set — lightweight and fast, documented and tested
STRUCTURE
- Full-stack Framework SvelteKit 2
- Reactive UI Svelte 5 runes
- Lint & Format Biome
DATA
- Relational DB PostgreSQL via Neon
- Graph DB Neo4j Aura
- Type-safe ORM Drizzle
- Object Storage Cloudflare R2
INTERFACE
- Atomic CSS UnoCSS
- Headless UI Bits UI
- 3D Rendering Three.js + Threlte
BEHAVIOR
- Session Auth Better Auth
- Form Validation Superforms + Valibot
- Compiled i18n Paraglide
INTELLIGENCE
- LLM Integration Vercel AI SDK
- Graph RAG Recursive retrieval
INSTANTIATE
compose from the full stack or a slice of it
v10r(landing-page)
uses partial stack
v10r(full-platform)
uses entire stack
v10r is not cloned — it's instantiated. An AI agent reads the tested patterns, architecture, and documentation, then emulates only the pieces the new project needs. The reference stays live. The instance is purpose-built.
PRACTICAL EXAMPLES
v10r(lynx)
uses SvelteKit· UnoCSS· Bits UI· markdown pipeline
dropped auth· databases· API· i18n· AI· 3D
keyboard firmware documentation — static prerendered
v4.lynxware.orgv10r(densho)
uses entire stack
lore-driven storytelling platform — entire stack, live
densho.mediaINTERNAL STRUCTURE
Composition and organization of the living system
Role-based agents with written souls — domain specialists that work as a team and run anywhere.
artyaesthetic refinement & visual polishdatydatabase schemas & data modelingdocydocumentation & technical writingresytechnology research & evaluationscoutreal-world usage researchsecysecurity review & threat modelingsveySvelteKit application patternstraydebugging & error tracinguxyUI/UX design & accessibility
agents/11 specialized Claude Code agentsskills/14 post-training knowledge modulesmemory/persistent agent memory across sessions
foundation/core concepts & conventionsstack/per-technology documentationblueprint/architecture decisionsimplementation/build detailspatterns/reusable patterns
Multi-client core — domain modules serve UI form actions, AI tool calls, REST API, and background jobs through a single business-logic layer.
Retrieval-Augmented Generation where the model chooses its own depth.
Two layers: a standard chunk index below, and above it an LLM-authored wiki with typed pointers into specific chunk IDs. The model reads TLDRs first and calls a drill-down tool only when it needs the raw source.
raw-RAGhybrid vector + BM25 retrieval over raw document chunksLLM-Wikisynthesized pages with typed pointers back to chunk IDsdrill-downtool-call escalation when the TLDR is not enough — budget 3 per turnverifypost-hoc citation taxonomy — quote, paraphrase, drifted, uncited