Cloudflare Developer Platform
Cloudflare's developer platform for building and deploying serverless applications and AI features on a global edge network. It bundles Workers AI inference, an agents runtime, an AI gateway, a vector database and edge storage behind one account.
What is Cloudflare Developer Platform?
Cloudflare Developer Platform is the build-and-deploy side of Cloudflare, documented at developers.cloudflare.com, which is where the address cloudflare.dev leads. It is not a single application but a catalogue of primitives running on the company's own network — 335 data centres in more than 125 countries — grouped on the front page into four families: compute, AI, storage and databases, and media.
The AI family is the part that concerns an AI directory most. Workers AI runs machine learning models on serverless GPUs: over fifty open-source models sit in a catalogue and can be called from a Worker binding, from the REST API, or through OpenAI-compatible endpoints, billed per Neuron instead of per rented GPU hour. AI Gateway stands in front of model calls, whether they go to Workers AI or to OpenAI, Anthropic and Google Gemini, adding analytics, logging, caching, rate limiting, retries, model fallback, guardrails in beta and prepaid unified billing. Agents provides a durable runtime in which every session keeps its own identity, local SQL storage, live connections, scheduled work and recoverable execution; agents answer over chat, email, voice, Slack or webhooks and reach for sandboxed code execution, browser control, search and Model Context Protocol tooling. Vectorize and AI Search handle embeddings, semantic search and retrieval-augmented generation over your own material.
Around that sit the parts an application actually needs. Workers and Containers cover compute; Durable Objects, Queues and Workflows handle coordination; R2, D1, KV and Hyperdrive hold data; Images, Stream and Realtime serve media. Security and networking — firewall, DDoS protection, TLS, bot management, Turnstile, Zero Trust, DNS, caching — live in the same account rather than in a separate product.
Everything is driven by code: a wrangler command line, a documented REST API, SDKs, Terraform and Pulumi providers, per-product llms.txt files and a setup page written for coding agents. A permanent free plan covers hobby projects and ten thousand Neurons a day, while the paid plan opens at a five-dollar monthly minimum and meters usage beyond its allotments, with no egress or bandwidth charges. There is no no-code route into any of it.
What it does
- Deploy serverless functions and full-stack applications to a global network with a single command
- Run inference on more than fifty open-source models without provisioning or scaling GPUs
- Host durable AI agents with their own identity, local SQL storage, sessions and recoverable execution
- Route, cache, rate-limit, retry and meter AI calls to several model providers through one gateway
- Index and query vectors for semantic search and retrieval-augmented generation over your own data
- Store and query data at the edge with object storage, serverless SQL, key-value and connection pooling
- Publish and consume Model Context Protocol servers, and drive an agent-first headless browser
When to use Cloudflare Developer Platform / When not to
A quick filter to help you decide if Cloudflare Developer Platform is the right fit.
When to use Cloudflare Developer Platform
- Full-stack developers who want to ship an application without provisioning or maintaining servers
- Engineering teams building AI agents that need durable sessions, scheduling and chat, voice, email or Slack channels
- Machine learning and platform engineers who need inference on demand without renting or scaling GPUs
- Site reliability and platform engineers looking for compute, storage, security and observability under one roof
- Technical founders and independent developers, whose hobby and side projects fit inside the permanent free plan
When not to use Cloudflare Developer Platform
- Non-technical users: every capability is reached through code, the wrangler CLI or the REST API, and there is no no-code interface
- Anyone looking for a ready-made chatbot or assistant, since this is a toolkit for building products rather than a finished one
- Teams that require dedicated GPUs or privately hosted custom models, which go through a custom requirements form and a sales conversation
- Organisations that need data to stay in one named jurisdiction by default, as the network is global and localisation is a separate add-on
- Anyone training large models, because the offering is serverless inference billed per Neuron rather than a training environment
- Buyers who need a flat, predictable monthly bill, since most of the cost is metered usage above plan allotments
How to use Cloudflare Developer Platform
A typical end-to-end flow, from setup to results.
- Create an account on the Cloudflare dashboard — the free plan needs no payment method
- Scaffold a project with a single command: npm create cloudflare@latest my-app
- Follow the Get started guide, or the Create your first Worker walkthrough, to deploy a first function
- Declare the resources your code needs as wrangler bindings: AI, KV, D1, R2, Vectorize or Durable Objects
- List the available models with npx wrangler ai models, then call one by binding, by REST API or through the OpenAI-compatible endpoints
- For an agent, start from the Agents SDK, pick a channel — chat, voice, email, Slack or webhook — and attach the sandbox, browser, search and MCP tools
- Put AI Gateway in front of your model calls to gain logs, caching, quotas and provider fallback
- Create a database when you need one: npx wrangler d1 create my-database
- Deploy, then watch behaviour through Analytics, Logs, Log Explorer and agent tracing
- Stay on the free plan until you hit its limits, then move to Workers Paid at a five-dollar monthly minimum
Pros & Cons
Pros
- A genuinely usable permanent free plan: Workers, Pages Functions, KV and Hyperdrive, plus ten thousand Neurons a day for AI
- No egress or bandwidth charges on the paid plan, which removes the usual surprise in cloud invoices
- Compute, storage, AI, security and content delivery come from one vendor and one account
- A written commitment that Customer Content is not used to train models or improve services without explicit consent
- An unusually deep compliance stack: ISO 27001, 27701 and 27018, SOC 2 Type II, PCI DSS Level 1, European Cloud Code of Conduct and C5 2020
- Dense documentation, versioned openly on GitHub, with llms.txt files and installation prompts aimed at coding agents
- Portability is respected: OpenAI-compatible endpoints, Model Context Protocol, Terraform and Pulumi providers
Cons
- The product surface is very wide, and the learning curve is steep for anyone new to the Workers execution model
- Usage-based billing is hard to forecast: Neurons, requests, CPU milliseconds and row reads each carry their own scale
- Several AI building blocks are still in beta, including the batch API, function calling, guardrails and the agent-first browser
- Cloudflare neither creates nor trains the models, so licence terms and behaviour depend on each third-party provider
- Data localisation is not the default: keeping traffic and logs in a chosen region requires the separate Data Localization Suite
- The developer documentation is English only, while the corporate site is published in thirteen locales
- There is no no-code path: everything assumes code, a command line or an API client
Pricing & Plans
A permanent free plan is available and requires no payment method: the Workers Free plan covers limited use of Workers, Pages Functions, Workers KV and Hyperdrive, and includes ten thousand Neurons per day of Workers AI inference. The lowest paid entry point is the Workers Paid plan, at a minimum charge of 5.00 USD per month per account, which raises the included allotments, meters usage beyond them, and carries no additional charge for data transfer or bandwidth. Workers AI usage above the free allocation is charged at 0.011 USD per 1,000 Neurons. The separate network and CDN plans start at 20.00 USD per month billed annually.
- limited use of Workers
- Pages Functions
- Workers KV and Hyperdrive
- plus 10
- 000 Neurons per day of Workers AI
- higher allotments
- metered usage beyond them
- Durable Objects included
- no egress or bandwidth fees
- Workers AI at 0.011 USD per 1
- 000 Neurons
- billed on the usage model set in the contract
- arranged through an account manager
- Free (network and CDN) — 0 USD per month
- aimed at personal or hobby projects that are not business-critical
- Pro (network and CDN) — 20 USD per month billed annually
- or 25 USD per month billed monthly
- Business (network and CDN) — 200 USD per month billed annually
- or 250 USD per month billed monthly
- Contract (network and CDN) — custom pricing
- billed annually
- for mission-critical applications
Data, GDPR & hosting
A consolidated view of how Cloudflare Developer Platform handles your data.
GDPR overview
GDPR implementation is documented rather than merely claimed. A dedicated page, titled "Cloudflare and GDPR compliance", sits in the Trust Hub, and a standard Data Processing Addendum is published and incorporated by reference into both the self-serve and the enterprise subscription agreements. It carries the EU Standard Contractual Clauses, alongside certification under the EU-U.S. Data Privacy Framework, its Swiss and UK extensions. Cloudflare holds ISO 27001, ISO 27701, ISO 27018, SOC 2 Type II and PCI DSS Level 1, and is certified to the European Cloud Code of Conduct and Germany's C5 2020 standard. A subprocessor list is public, a Data Protection Officer is designated at dpo@cloudflare.com, and rights requests go to sar@cloudflare.com with a thirty-day response commitment. No Article 27 EU representative is named on the pages reviewed.
Who owns the data?
Under the Self-Serve Subscription Agreement, you and your end users keep all right, title and interest in the Customer Content you send through the services — data, code, images, video, prompts, model outputs and embeddings alike. Cloudflare takes only a non-exclusive, sublicensable, worldwide and royalty-free licence, and expressly limits it to what is necessary to deliver the services. The line is drawn elsewhere: Network Data, meaning the models, statistics, analyses and databases Cloudflare derives from traffic on its own network, belongs to Cloudflare, which may use it to run and improve its services. Content posted on the public Cloudflare websites falls under a separate and far broader perpetual, irrevocable licence.
Reuse rights
You can reuse your own content and the outputs you generate without asking Cloudflare for permission: nothing in the subscription agreement conditions their reuse, and no revenue share or attribution is imposed. Two caveats deserve attention. Cloudflare neither builds nor trains the models offered through Workers AI — they are third-party services carrying their own open source or commercial licence terms, which bind you directly to each model provider and should be checked before commercial use. And you remain responsible for holding the rights to whatever you transmit. Cloudflare, for its part, commits not to use Customer Content to train the models or improve its services without your explicit consent, and never exposes it to another customer. In the other direction, the website terms forbid scraping Cloudflare's own pages with bots to train an AI system unless robots.txt allows that user agent.
Data retention & training
Hosting summary
Cloudflare runs its own network: edge servers in more than 330 cities, 335 data centres, over 125 countries. For its core application layer services the bulk of the data stays on those edge servers, and Cloudflare states that it does not store customer content for them; metadata about that activity, and log data containing limited personal information such as IP addresses, are processed in data centres in the United States and Europe for a limited period. Certain products do store data by design — R2, KV, D1, Durable Objects and Vectorize among them. Where the data sits is configurable rather than fixed: the Data Localization Suite combines Regional Services and Geo Key Manager, which control where TLS keys are held and where traffic is inspected, with the Customer Metadata Boundary, which controls where logs are sent. Transfers from the EEA to the United States rest on the Standard Contractual Clauses in the Data Processing Addendum and on Data Privacy Framework certification. The company is headquartered in the United States and keeps five European offices.
Where Cloudflare Developer Platform works
Country-level availability.
Not available in
Things to keep in mind
Risks and trade-offs to weigh before adopting Cloudflare Developer Platform.
- Metered billing rewards careless code: an unbounded agent loop can burn through Neurons and requests without an obvious ceiling
- Concentrating compute, data, security and AI with one vendor makes the platform very hard to leave later
- The models are third-party services, so their licences, their biases and their availability are outside Cloudflare's control
- Beta components carry no obligation to fix bugs and may be withdrawn or left unreleased at any time
- Network Data derived from your traffic remains Cloudflare's property and is used to improve its services
- Model output can be wrong, and the agreement is explicitly as-is with all warranties disclaimed — human review stays your responsibility
- An agent that drives a browser, executes code in a sandbox and touches payments widens the attack surface and needs deliberate scoping
Setup & Integrations
Technical difficulty
Moderate for a developer, out of reach for anyone else. A free account and one command — npm create cloudflare@latest my-app — put a working project on the network within minutes, and the documentation carries step-by-step guides for each product plus setup prompts written for coding agents. It nonetheless assumes Node.js, a command line, and the notion of bindings between code and resources. The advanced pieces — Durable Objects, Workflows, Vectorize, Model Context Protocol servers — call for real backend experience. There is no visual builder and no no-code entry point of any kind.
Deployment
Integrations
Supported languages
Behind Cloudflare Developer Platform
Fundraising
Social
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
Is cloudflare.dev a separate product?
Is there a free plan?
What does the first paid tier cost?
Is my data used to train AI models?
Does Cloudflare build the models it offers?
Is there an API?
Where is data processed?
Is a Data Processing Agreement available?
What certifications does Cloudflare hold?
What is the minimum age to use the services?
Should you pick Cloudflare Developer Platform?
Cloudflare Developer Platform is one of the few AI-capable platforms whose maturity is not in question: it is published by a company listed on the New York Stock Exchange since 2019, running its own network of 335 data centres, and its AI catalogue has been assembled on top of infrastructure that already existed rather than the other way round. That shows in what the platform is good at. Inference, agents, vector search and gateway control sit next to compute, storage, security and content delivery in a single account, with no egress or bandwidth charges and a free tier substantial enough to carry a real side project.
Two commitments stand out for anyone weighing an AI vendor. Customer Content is contractually excluded from model training and service improvement without explicit consent, and the compliance stack — ISO 27001, 27701 and 27018, SOC 2 Type II, PCI DSS Level 1, the European Cloud Code of Conduct, Germany's C5 2020, a published Data Processing Addendum with Standard Contractual Clauses and a public subprocessor list — is deeper than almost anything else in this directory.
The reservations are practical rather than structural. Metered pricing across Neurons, requests, CPU time and storage operations is genuinely hard to forecast before the first invoice, and an unbounded agent loop can spend real money quietly. Several AI components are still in beta, with the weaker guarantees that implies. The models themselves are third-party, so their licences and their behaviour are not Cloudflare's to promise. Data localisation costs extra. And nothing here is reachable without code.
For developers, platform and reliability engineers, and technical founders, it is a serious and well-documented option. For anyone hoping to assemble an AI product without writing any, it is the wrong tool entirely.
- Choosing a selection results in a full page refresh.
- Opens in a new window.