Deploy in 5 minutes
npm install -g wrangler wrangler login
git clone https://github.com/SuperInstance/quilt-cloudflare cd quilt-cloudflare npm install
wrangler d1 create quilt-db wrangler vectorize create quilt-embeddings --dimensions=768 wrangler kv:namespace create CACHE
[[d1_databases]] binding = "DB" database_name = "quilt-db" database_id = "..." # from step 3 [[vectorize]] binding = "VECTORIZE" index_name = "quilt-embeddings" [[kv_namespaces]] binding = "CACHE" id = "..." # from step 3 [ai] binding = "AI"
wrangler d1 execute quilt-db --file=./schema.sql wrangler deploy # → https://quilt-cloudflare.YOUR_SUBDOMAIN.workers.dev # 🎉 Your Quilt is live on the edge.
What you get
A full Quilt runtime on Cloudflare's edge, with all the bells and whistles.
The reactive engine
Every cell is a real Quilt cell. Reactive propagation. Per-context memoization. Topological evaluation.
D1 persistent state
SQLite at the edge. Every cell value is persisted. Full history. Multi-region read replicas.
Vectorize semantic search
Embed every cell value. Search semantically. Find the right cell by meaning, not just name.
KV cache
Fast ephemeral state in 300+ cities. Formulas, listeners, intermediate computations.
R2 backups
Versioned backups in object storage. Full sheet snapshots. Time travel via R2.
Workers AI
LLM cells. Embedding cells. Image cells. Translation. Sentiment. Summarization. Code generation.
MCP server
Every cell is an MCP tool. Plug into Claude Code, Cursor, or any MCP client.
Pages UI
A full visual editor deployed as a Page. Studio, snippets, semantic search. Free SSL.
7 AI cells, built in
Every Workers AI capability is a Quilt cell. Same addressable model. Same reactive propagation.
Call any text-generation model.
- id: explain kind: ai.llm model: "@cf/meta/llama-3-8b-instruct" prompt: '"Explain: " + topic'
Generate 768-dim embeddings. Index in Vectorize.
- id: vec kind: ai.embed input: "text"
Image classification with a pre-trained model.
- id: classify kind: ai.image input: image.url
Translate between 100+ languages.
- id: es kind: ai.translate from: "en" to: "es" input: "text"
Sentiment score 0-1 for any text.
- id: score kind: ai.sentiment input: "text"
Summarize long text into a few sentences.
- id: summary kind: ai.summarize input: "long_text"
Generate code from a prompt.
- id: gen kind: ai.code language: "python" prompt: "Compute fib"
Transcribe audio to text.
- id: transcript kind: ai.speech input: audio.url
Architecture
Three layers, all on the edge. The Quilt engine runs in a Worker. State lives in D1. The UI is a Page.
┌─────────────────────────────────────────────┐
│ Cloudflare Edge (300+ cities) │
│ │
┌─────────┐ │ ┌──────────────────────────────────────┐ │
│ Browser │◀────▶│ │ Worker (the Quilt engine) │ │
│ UI │ │ │ │ │
└────┬────┘ │ │ ┌────────────┐ ┌─────────────┐ │ │
│ │ │ │ Reactive │ │ AI cells │ │ │
│ │ │ │ engine │──▶│ (LLM, etc) │ │ │
│ │ │ └─────┬──────┘ └──────┬──────┘ │ │
│ │ └─────────┼────────────────┼──────────┘ │
│ │ │ │ │
│ │ ┌─────────▼──────┐ ┌──────▼──────┐ │
│ │ │ D1 │ │ Vectorize │ │
│ │ │ (cell state) │ │ (semantic) │ │
│ │ └────────────────┘ └─────────────┘ │
│ │ ┌────────────────┐ ┌─────────────┐ │
│ │ │ KV │ │ R2 │ │
│ │ │ (cache) │ │ (backups) │ │
│ │ └────────────────┘ └─────────────┘ │
│ │ │
│ │ ┌──────────────────────────────────────┐ │
│ │ │ Pages (the web UI) │ │
│ │ │ Studio / Live / Playground │ │
│ │ └──────────────────────────────────────┘ │
│ │ │
└───────────┴─────────────────────────────────────────────┘
What to build on it
Memory as cells. Tools as cells. Reasoning as cells. The whole agent in a sheet, running on the edge, accessible via MCP from any AI client.
Sensor cells, formula cells, listener cells. WebSockets push updates to the UI. D1 persists the history. Vectorize makes the data searchable.
One Quilt deployment, many sheets. Each sheet is a tenant. Router cells enforce per-tenant policies. Vault cells encrypt the sensitive data.
ESP32 sensors push to a Cloudflare Worker. The worker routes through Quilt cells. D1 persists. Vectorize indexes. Workers AI summarizes. The whole pipeline, on the edge.
Add notes, they auto-embed. Ask a question, the most relevant cells are retrieved and sent to the LLM. RAG over your personal data, with no third-party.
The HTTP API
Every cell addressable over HTTP. The sheet is the API surface.
# Load a sheet
POST /sheet?id=my-app
Content-Type: text/plain
<YAML sheet here>
# List all sheets
GET /sheets
→ { "sheets": ["default", "my-app"] }
# Get a cell
GET /cell/heat_index
→ { "data": 24.5, "status": "ready", "t": 5, "author": "cloudflare" }
# Set a cell
POST /set/heat_index
Content-Type: application/json
{ "value": 30 }
→ { "ok": true }
# Get all cells in the current sheet
GET /cells
→ { "id1": { "data": ..., "status": "ready" }, ... }
# MCP over HTTP
POST /mcp
{ "jsonrpc": "2.0", "id": 1, "method": "tools/list" }
# MCP over SSE
GET /mcp/sse
What it costs
Cloudflare's free tier is generous. Paid plans scale smoothly.
┌──────────────────┬────────────────┬─────────────────┐ │ Resource │ Free tier │ Paid plan │ ├──────────────────┼────────────────┼─────────────────┤ │ Worker requests │ 100K/day │ 10M+ included │ │ Worker CPU time │ 10ms/req │ 30s/req │ │ D1 reads │ 5M/day │ Unlimited │ │ D1 writes │ 100K/day │ 50M+ included │ │ D1 storage │ 5GB │ 10GB+ │ │ Vectorize queries │ 30M/month │ Unlimited │ │ KV reads │ 100K/day │ Unlimited │ │ Workers AI │ 30M neurons/day│ Pay as you go │ │ Pages bandwidth │ Unlimited │ Unlimited │ └──────────────────┴────────────────┴─────────────────┘ For most personal use, free tier is more than enough. For production, ~$5/mo covers a small team's Quilt.