sSystm ships with AI inside — an editor AI grounded in your brand, your components and your skills. And the whole platform is an MCP surface for the AI you already work with. Same context, both sides — and you stay in control.
The real architecture: one context bundle feeds the built-in AI and your own — verified against the code.
Not a chatbot in a corner. An editor AI that writes in your brand, an assistant that answers from your workspace — running on Cloudflare, billed through Cloudflare.
The editor AI in Build streams real code, file by file — on Anthropic models through Cloudflare AI Gateway, with prompt caching cutting repeat costs. You choose the model.
A built-in assistant on Workers AI (Llama 3.3 70B through AI Gateway) that answers grounded in a live bundle of your workspace — and is instructed to never invent data, numbers or names.
Wireframe generation runs on Workers AI too, with a Cloudflare Workflow carrying the long jobs.
Workers AI for the assistant and embeddings, AI Gateway in front of every model call, Vectorize for semantic search, Durable Objects for real-time — and team voice in Chat on the Realtime SFU.
No second AI vendor to onboard. Usage is billed through Cloudflare, with per-org cost metadata on every call.
Everything in sSystm is connected to everything — so before the Build AI writes a line, the platform assembles the truth around it. Four layers, straight from the code:
Claude on claude.ai, Claude Code in your terminal, Claude Desktop, Cursor — any MCP client. One endpoint, one token, and your AI works inside your workspace.
sstm_* — real writes: deals, projects, tokens, components, wireframes, skills, chat
sstm:// — your live workspace as structured, readable data
component · wireframe · design — the exact data models your AI builds with
/api/mcp/stream — Streamable HTTP with a per-user Bearer token
sstm://project/{id}/contextEverything an AI needs to build for one project, bundled: the project, its active design system — tokens, rules and the resolved brand CSS — linked components with full code, attached skills with their instructions and template files, and the wireframes. The same bundle the built-in Build AI is grounded in.
sstm://component/recipesstm://wireframe/recipesstm://design/recipesSystm never generates on your AI's behalf. It serves recipes — the full token vocabulary, the wireframe data model with all 75 blocks, worked examples. The server's own words: "YOU write the component yourself."
The exact setup the MCP module in the app walks you through — endpoint /api/mcp/stream, one Bearer token per user.
claude mcp add --transport http sstm \
https://<your-workspace>/api/mcp/stream \
--header "Authorization: Bearer <your-token>"{
"mcpServers": {
"sstm": {
"url": "https://<your-workspace>/api/mcp/stream",
"headers": { "Authorization": "Bearer <your-token>" }
}
}
}{
"mcpServers": {
"sstm": {
"command": "npx",
"args": ["-y", "mcp-remote",
"https://<your-workspace>/api/mcp/stream",
"--header", "Authorization: Bearer <your-token>"]
}
}
}Settings → Connectors → Add custom connector
https://<your-workspace>/api/mcp/stream?token=<your-token>Every write your AI makes lands visibly in the workspace — components render live in the catalogue, chat posts are always attributed as AI, reviews reach your Health view as findings. And infrastructure changes on your Cloudflare account are gated: safe reads may run, anything risky is staged as pending until a human approves it.
AI-written components, wireframes and messages appear instantly in the workspace, attributed and reviewable by the team.
sstm_cloudflare_execute never auto-runs risky or destructive operations — they wait as pending approvals in the UI.
Your AI records findings on real project output with severities — and you decide what gets resolved.
Join the waitlist — when your spot opens, the built-in AI writes in your brand from day one, and Claude, Cursor or any MCP client connects to the same workspace, gated by you.
When you're in: no email, no password — one Cloudflare grant · core workspace free · your data lives in your own account