NIMOS / ENGINES / DILAISDK
← THE BUILD SYSTEMDILAISDK
THE DESIGN BRAIN
Design Intelligence Layering & AI SDKs
The layer that turns market research into structured design decisions, not hallucinated HTML.
DRAWN BY THE HOUSE
One mark per engine, on one grid, in one hand.
11
TOOLS
PROJECT
WIRE
USED
THIS BUILD
A live model call site in lib/cloud-research.ts.
THE PROOF LEDGER →WHAT
What it is
Eleven tools: Vercel AI SDK with OpenAI and Anthropic adapters, Google GenAI, LangChain JS, Pinecone, Zod, CopilotKit, Trigger.dev, Hugging Face inference and zod-to-json-schema.
WHY
Why it exists
Generated UI fails when it is unstructured. Schemas keep model output safe; retrieval grounds it in the client's actual market; streaming makes it usable instead of a spinner.
HOW
How it runs
npm into projects; keys read at runtime from the standing lanes; Python/agent lanes share the same contracts. Every LLM path is schema-tested before it reaches a page.
WHEN
When it runs
AI features in a build, structured outputs, retrieval/context injection, copilot bindings and background generation jobs.
WHO
Who owns it
CEO owns the intelligence layer and cost altitude; Research watches providers; Build wires streaming UI; Data tracks schemas, indexes and tokens; Ops tests every path.
TOOLS
The registry
The engine manifest is the source of truth. These are its entries, including the ones that are not used on this build — the proof ledger says which is which.
ai
npmVercel AI SDK — stream text, objects and structured UI from LLMs into Next.js views.
@ai-sdk/openai
npmOpenAI provider adapter with structured JSON schema output.
@ai-sdk/anthropic
npmAnthropic provider adapter — high-fidelity reasoning over raw text layouts.
@google/genai
npmLow-latency multimodal context ingestion — video/image parsing.
langchain
npmModular framework for compound sequential agent tasking.
@pinecone-database/pinecone
npmManaged vector index for injecting workspace context.
zod
npmRuntime TypeScript-first schema validation keeping dynamic AI output safely typed.
@copilotkit/react-core
npmReact hooks giving LLMs direct authority over local UI state.
@trigger.dev/sdk
npmHeavy asynchronous job coordination for Next.js and Vercel.
@huggingface/inference
npmDirect client/server model loading for targeted embedded vision nodes.
zod-to-json-schema
npmConverts Zod validation boundaries into JSON Schema templates for structured LLM output.
11 registry entries. The proof ledger on the Gantry page says which of them this build actually used.
RUN
Run it
ENGCTL / THE ENGINE FLOOR
$ engctl show dilaisdk$ engctl awaken dilaisdk$ engctl install dilaisdk --dir ./site`engctl` is the registry control surface. `awaken` checks wiring before it activates a cell; `install` writes the parts into a project.
PROOF
On this build
A live model call site in lib/cloud-research.ts.
N/A and PARTIAL are stated, not hidden. The ledger is generated from this project, not from a template.
LIMITS
What this does not claim
This site uses one live model call site in lib/cloud-research.ts. The rest of the engine is available, not claimed as used.
The engine is the how. The brief is the start.