Frequently asked questions

SwiftTune is a visual platform for composing, deploying and improving AI applications that run on Cloudflare. You drag blocks onto a canvas, wire them into a flow, and publish that flow to a live edge URL. The platform owns the plumbing between the Cloudflare services so you can spend the time on the product instead.

A flow is a graph: every block has typed input and output ports, and a connection is only valid when the two ends agree. The graph is validated before it can be saved, so unknown blocks, mismatched ports and cycles are rejected while you are still on the canvas rather than at run time.

At run time the engine walks the graph in dependency order and hands each block the output of the ones before it. See Flow and Block in the glossary.

The block catalogue covers the services an AI application usually needs:

• Workers AI for inference on Cloudflare-hosted models.
• AI Gateway for multi-provider routing, caching and retries.
• Vectorize for embeddings storage and similarity search.
• R2 for blobs and artefacts, and Queues for work that outlives a request.
• Agents, AI Search, MCP and session memory for stateful, tool-using flows.
• HTTP for anything else you already run behind a URL.

No. A working flow is composed on the canvas and configured through the properties panel — no SDK wiring and no deployment scripts. Blocks that call your own services take a URL and a schema, so you can bring existing code in without rewriting it as part of the platform.

Deploying publishes the current version of the flow as a Cloudflare Worker and returns the URL it answers on. The publish runs asynchronously, so the request does not block on it, and each deployment keeps the version it came from. If a release misbehaves you can roll back to the previous version instead of redeploying by hand.

Yes — flows run in your account, which is what keeps the underlying Cloudflare usage billed to you at cost and your data inside a boundary you control. SwiftTune stores the configuration and the telemetry it needs to operate, not your payloads. See Bring your own account.

Yes. Every persistent store the platform creates — D1, R2 and any other durable storage — is provisioned with the EU jurisdiction flag, which is what pins the data to European data centres rather than merely naming it so. Personal data is never written to the global key-value cache. The Privacy Policy states the sub-processors and the transfer position.

No end-user payload is kept at rest. The platform records the flow configuration, the deployment history and operational telemetry; the prompts, documents and responses that pass through a running flow stay in your account. The Data Processing Agreement sets this out in Article 28 terms.

The quality add-ons watch the flows you already shipped: evaluations score a flow against a fixed dataset, retrieval monitoring reports on the documents your searches actually return, and cost attribution breaks spend down per flow and per block.

Because the analysis is replayed from recorded runs rather than from wall-clock time, the same input produces the same verdict twice. See Evaluation and Retrieval-augmented generation.

A published flow sits behind Cloudflare’s WAF and DDoS protection, and every tenant carries a rate limit matched to its plan — enforced at the edge and again in the runtime, so a burst is rejected with a canonical 429 before it reaches a model. The Safety Shield block adds policy checks inline in the flow itself.

Plans run from a free tier through Starter, Team, Pro and Scale, with Enterprise priced on request. Each tier raises the throughput and unlocks more of the quality add-ons, which can also be bought on their own. The Cloudflare resources your flows consume are billed by Cloudflare to your own account.

Compose flows, deploy them and run them — no card required. The free tier is limited on throughput and on the quality add-ons, not on the composer: the blocks and the deployment path are the same ones the paid tiers use.

Sign-in goes through Auth0 with email or Google. SwiftTune never sees or stores a password: the identity provider is the only authority on who you are, and the platform checks the token it issues on every request.

No. SwiftTune is an application platform, not a model lab: it composes the models Cloudflare hosts and the providers you route to through AI Gateway. Nothing you send through a flow is used to train a model.