AI application platform
Compose AI applications on Cloudflare in minutes.
Visual builder for Workers AI, Gateway, Vectorize and more. Deploy globally, monitor with Evals and Copilot, pay only for what you run.
Free tier, no card. Bring your own Cloudflare account.
Cloudflare, already wired
- Workers AI
- AI Gateway
- Vectorize
- AI Search
- Agents
- R2
- Queues
- MCP
- Session Memory
Shipping an AI application on Cloudflare still costs weeks
The primitives are excellent. Assembling them into something you can put in front of users is where the time goes.
Weeks of glue before the first user
Workers AI, Vectorize, AI Gateway, R2 and Queues each ship their own primitives. Turning them into something production-ready is infrastructure work that delays the product itself.
Every tool solves one slice
Model hosts do not evaluate, evaluation platforms do not host, and workflow builders are not AI-native. Nothing closes the loop from compose to deploy to observe to improve.
Regressions surface in production
Prompt regressions, retrieval drift and runaway spend show up after they reach users, because no single system watches the flow end to end.
Compose it, deploy it, then keep it honest
One canvas from the first block to the regression you caught in production. No handover between four vendors, because there is no handover.
Compose
Drag the blocks, wire the ports
Every Cloudflare service an AI application needs is already a block. The graph is validated as you draw it, so unknown blocks, mismatched ports and cycles are rejected on the canvas rather than at run time.
27 blocks in the catalogue
Deploy
One click to a live edge URL
The flow publishes to Cloudflare with WAF, DDoS protection and per-tenant rate limiting already in front of it. No infrastructure to stand up, no pipeline to write first.
Sub-100ms, globally
Tune
Catch it before your users do
Evals, Traces, Retrieval Monitor and Cost Attribution report on the flows you shipped, and Copilot watches all of them for you. An improvement starts from evidence instead of a hunch.
Evidence, not hunches
Run it before you deploy it
Type a question and run the flow below. Each block reports what actually happened — including the ones that only need a deployment to go further.
- HTTP TriggerNot run yet.
- AI SearchNot run yet.
- AI GatewayNot run yet.
- ResponseNot run yet.
The part that tells you it stopped working
Shipping the flow is the easy half. These are the modules that watch it afterwards — each sold on its own, each bundled from the Team tier up.
What the platform commits to
- EU
- Data residency on D1 and R2, by jurisdiction
- Zero
- End-user payloads stored at rest
- BYO
- Your Cloudflare account, billed to you
- 12
- Modules, each sold on its own
Where the first customer stories will go
We are in early access. Nothing here is a customer we have not met.
Reserved for an early customer
A quote or a logo lands here the day a real customer gives us one, not before.
One price for the platform. Cloudflare stays yours.
Every tier runs on the Cloudflare account you already own, billed to you at Cloudflare's rates. What you pay us is the platform on top of it.
Free
$0/mo
One flow, no card, no expiry.
Starter
$19/mo
A side project that has users.
Team
Popular$99/mo
The Quality suite bundled, for a product in production.
Pro
$399/mo
Every module, higher throughput, priority support.
Enterprise
Custom
From $1,499 at Scale: SSO, dedicated support, contracted terms.
Add-ons are sold standalone too, so a tier is a starting point and not a ceiling. See what each tier includes.
Comparing us with something else?
Five comparisons, each one naming what the other product does better and how to move across if you still want to.
- SwiftTune vs Braintrust
- Stay on Braintrust if evaluation is the work and the application is already built and hosted the way you want it.
- SwiftTune vs Humanloop
- Stay on Humanloop if the people who improve your prompts are not the people who deploy your application.
- SwiftTune vs LangSmith
- Stay on LangSmith if the application is LangChain or LangGraph and you want the tracing that was built for it.
- SwiftTune vs Vercel
- Stay on Vercel if the AI feature is one part of a web product you are already shipping there.
- SwiftTune vs n8n
- Stay on n8n if the workflows move data between SaaS tools, or if self-hosting is not negotiable.
Your first flow is fifteen minutes away
Drag the blocks, wire the ports, deploy to a live URL. Nothing to stand up first, and nothing to take down if you change your mind.
Free tier, no card, no expiry.