z.ai GLM 5.3 as a Quilt cell

Cell: ai.complete

Plain text completion. The most basic cell. Just a prompt, the model responds.

Write a tagline for a personal data mesh that runs on every device.

Cell: ai.code

Generate code from a description. The cell returns just the code.

a function that returns the n-th Fibonacci number, with memoization

Cell: ai.reason

Reason step-by-step. The cell returns both reasoning and a final answer.

A farmer has 17 sheep. All but 9 run away. How many are left?

Cell: ai.summarize

Summarize a long text into a short one.

Quilt is a reactive, typed, cellular runtime. It takes the spreadsheet model — a grid of named, reactive cells — and turns it into a general-purpose runtime. A cell can be a value, a formula, a program, a sensor, an API, a listener, a router, or an IO port. The same sheet can be deployed to a browser, a server, an edge function, a microcontroller, or a peer-to-peer mesh. The cell model is uniform across all of them.

Cell: ai.qa (RAG-style)

Answer a question given context. The cell uses the context to ground its answer.

Quilt has 8 cell kinds: value, formula, program, sensor, api, listener, router, io. Each is a first-class primitive. The runtime handles propagation automatically. Sheets are YAML files. The same sheet runs in TS, Rust, and the browser. What are the 8 cell kinds?

Cell chain: 3 cells in series

The output of each cell feeds the next. A reactive cascade across 3 LLM calls.

A new feature: Quilt cells can be encrypted end-to-end. The server only sees ciphertext.

Cell 1: extract key points → Cell 2: turn into tweet → Cell 3: explain to a 5-year-old