Cloudflare Workflows

Cloudflare Workflows is a cloud service for developers building durable, multi-step applications on Cloudflare Workers. It is especially suitable for teams already using Workers that need retries, persistent state, long pauses, or approval steps. Its durable execution and CPU-free waiting are compelling, but usage charges and a Workers for Platforms restriction limit where it belongs.

Overview

Workflows coordinates application steps without requiring you to operate orchestration infrastructure. Failed tasks can retry automatically, and state can persist for minutes, hours, or weeks. That makes it a good fit for processes that must survive a failed request, delayed dependency, or temporary outage rather than finish in one uninterrupted invocation.

Waiting does not consume CPU time, which can make long pauses economical. Billing still applies to requests, CPU time, persisted storage, and executed steps, so busy or finely divided workflows require cost monitoring. Workflows also cannot be deployed to Workers for Platforms namespaces, ruling it out for that deployment model.

Key features

• Durable steps: Automatic retries and error handling reduce the recovery code your application must own. Persisted state supports genuinely long-running processes.
• Scheduled waiting: step.sleep() and step.sleepUntil() pause execution for seconds, hours, or days without using CPU time while asleep.
• External events and approvals: A workflow can wait for webhooks, user input, or an external system response, making it suitable for human tasks and approval gates as well as automated jobs.
• Programmatic control: Instances can be triggered, paused, resumed, or terminated through code or an API. Invocation is available through the Workers API, Wrangler CLI, or REST API.
• Cloudflare integrations: Workflows can make Fetch API requests and call Cloudflare services such as R2, KV, and D1, keeping orchestration close to Workers-based applications.
• Scale and lifecycle: Up to 50,000 concurrent runs, workflow versioning, and unlimited wall-clock duration per step support substantial and evolving workloads. CPU time per invocation remains bounded by the configured limit.

Pricing

Cloudflare uses a freemium model, with paid service advertised from $5/mo. The free Workers Free plan costs $0.00 and shares 100,000 requests per day with Workers. It includes 10 ms CPU per invocation, 1 GB-month of storage, 3,000 steps per day, 100 MB of persisted state per instance, and 1,024 steps per Workflow. Those caps suit prototypes and modest automations, but daily requests and steps can become constraints quickly.

Workers Paid has custom pricing, with usage-based charges for requests, CPU time, storage, and steps. It includes 10 million requests per month, then charges $0.30 per additional million, and 30 million CPU milliseconds per month, then $0.02 per additional million. The plan is the more credible choice for production workloads, provided you can forecast step volume and persisted data. Sleeping remains CPU-free, but it does not eliminate request, storage, or step charges.

Platforms

Workflows is delivered as a cloud service through API and web tooling. It is designed for Cloudflare Workers deployments and can be managed with the Workers API, Wrangler CLI, or REST API. It does not support Workers for Platforms namespaces, a decisive limitation for multi-tenant platform deployments.

Who it's for

Choose Workflows if your application already runs on Workers and needs durable retries, delayed execution, external callbacks, or human approvals. It is well matched to provisioning, asynchronous processing, scheduled follow-ups, and other long-running workflows that benefit from Cloudflare services such as R2, KV, or D1. It is a weaker fit for teams seeking a self-hosted orchestrator, broad non-Cloudflare deployment, or predictable flat pricing.

Pros and cons

• Pro: Automatic retries, error handling, and persisted state cover failure and delay without a separate orchestration system.
• Pro: Sleeping and waiting consume no CPU time, improving the economics of long pauses.
• Pro: Webhooks, user input, and API lifecycle controls support workflows that include people and external systems.
• Con: Usage-based billing spans requests, CPU, storage, and steps, making costs less predictable than a simple subscription.
• Con: The free tier has shared daily request, CPU, storage, step, and state caps that can restrict growing workloads.
• Con: Workers for Platforms namespaces are unsupported, excluding that deployment route.

Alternatives

For a broader process-orchestration catalog, see Process Orchestration Software.

• Joget is a freemium, multi-platform option with a free tier for three users and an unlimited-workflow Starter plan.
• Bonita offers a free plan with 150 process instances per month and no time limit, across web, desktop, API, and self-hosted deployments.
• Camunda provides free self-managed development and testing with production usage listed separately.
• Dapr is free for development and non-critical workloads, with durable workflows and native integrations.
• Bizagi Modeler suits process modeling, with a free one-user plan and local storage.
• Cadence is open source with no license key, usage limits, or per-execution fee.
• Hatchet offers pay-as-you-go usage with 1 million runs and 1 million events per month free.
• Inngest has a free plan with five seats, 50,000 executions, and 500,000 ingested events.

Verdict

Cloudflare Workflows is a strong choice for Workers developers whose applications need durable, retryable steps and long waits, especially when Cloudflare data services are part of the design. Its main reason to choose it is durable execution without CPU cost during waiting. Look elsewhere if you need Workers for Platforms deployment, self-hosting, or a simpler fixed-price model.