COST ARCHITECTURE
See what can launch without immediate software overhead.
A launch allowance is a provider’s documented allocation or credit in a client-owned account. It can reduce early operating cost. It does not reduce the strategy, design, engineering, security, testing, or verification required to build the system.
| Capability and provider | Current allowance | Expected light use | What consumes it | Fallback | Upgrade path | Verification |
|---|---|---|---|---|---|---|
| Serverless application logic Cloudflare Workers recurringproduction-approved | 100,000 Worker requests per day; static asset requests should bypass Worker execution. Commercial use permitted under provider terms. | A static-first marketing site with low-volume forms and optional module requests. | Dynamic Worker or Pages Functions invocations. Static asset requests do not consume this request allocation. Review logs and binding data; do not send unnecessary sensitive data. | Static content and conventional navigation remain available. | Dynamic requests approach 70% of the daily allocation or CPU limits constrain execution. Likely paid successor: Cloudflare Workers Paid | approved-production |
| Spam and abuse protection Cloudflare Turnstile recurringproduction-approved | Provider-managed free plan; server-side token validation required. Commercial use permitted. | Abuse protection for contact and project-brief forms at typical small-business traffic. | Widget challenges and token verification requests under the provider-managed free plan. Document privacy disclosure and validate tokens server-side. | Honeypot, rate limit, input constraints, and manual contact path. | Enterprise policy, support, or specialized controls are required. Likely paid successor: Cloudflare enterprise agreement where required | approved-production |
| Transactional email Resend recurringproduction-approved | 3,000 emails/month, 100/day, one custom domain, one webhook endpoint, 30-day retention. Commercial use permitted subject to current terms. | Transactional visitor confirmations and owner notifications for a low volume of qualified inquiries. | Every sent or received email counts toward the daily and monthly quota. Avoid sensitive content in email; redact logs. | Store the structured record and show a manual follow-up path; retry through queue. | Daily or monthly volume approaches 75%, longer retention is needed, or multiple domains are required. Likely paid successor: Resend paid plan or alternate email adapter | approved-production |
| Structured data Cloudflare D1 recurringproduction-approved | 5M rows read/day, 100K rows written/day, 5 GB account storage, up to 10 free databases, 500 MB/database, seven-day point-in-time recovery. Commercial use permitted. | Structured leads, consent records, content metadata, and low-volume module status. | Rows read or scanned, rows written, and stored database data; unindexed scans consume more rows. No payment cards, passwords, government identifiers, or sensitive records by default. | Queue writes, retain a minimal encrypted export, or degrade to email-only intake. | Row scans, storage, recovery, or support requirements exceed the allocation. Likely paid successor: Cloudflare D1 paid usage or approved SQL provider | approved-production |
| Object storage Cloudflare R2 recurringproduction-approved | 10 GB-month, 1M Class A operations, 10M Class B operations monthly; no R2 internet-egress charge; excludes Infrequent Access. Commercial use permitted. | Approved downloads, monitoring screenshots, and a limited library of client-owned media. | Standard-storage GB-months plus Class A mutation and Class B read operations. Private buckets by default; signed access and retention controls required. | Disable optional uploads and preserve local/documented delivery path. | Storage or operation thresholds approach 75%, or security requirements need another service. Likely paid successor: Cloudflare R2 paid usage or approved storage adapter | approved-production |
| Read-heavy state and cache Cloudflare Workers KV recurringproduction-approved | 100K reads/day, 1K writes/day, 1K deletes/day, 1K lists/day, 1 GB storage. Commercial use permitted. | Feature flags, public configuration, redirect maps, and reusable cached results. | Key reads, writes, deletes, list operations, and stored data. Eventually consistent; not for critical transactional records. | Read from typed repository or D1 source of truth. | Write volume or consistency requirements exceed KV's design. Likely paid successor: Workers KV paid usage or D1 | approved-production |
| Asynchronous workflow Cloudflare Queues recurringproduction-approved | 10,000 operations/day and 24-hour message retention; a typical successful delivery uses three operations. Commercial use permitted. | Low-volume email delivery, indexing, report creation, and retryable background work. | Each 64 KB write, read, and delete operation; a typical successful message consumes three operations. Use idempotency, retry limits, and dead-letter behavior. | Process synchronously only when safe, or store a retryable pending record. | Operations approach 75%, retention is insufficient, or higher guarantees are required. Likely paid successor: Cloudflare Queues paid usage | approved-production |
| AI governance and routing Cloudflare AI Gateway recurringproduction-approved | Core features free; up to 10 gateways and 100,000 stored logs across the account. Commercial use permitted. | Governance, logging, caching, and limits for low-volume optional AI features. | Stored logs across gateways; model inference and guardrail usage are accounted for separately. Redact prompts and avoid sensitive data in unpaid providers. | Deterministic workflow, cached response, conventional search, or contact path. | Log volume, policy, support, or enterprise governance exceeds the free plan. Likely paid successor: Cloudflare paid/enterprise services or provider-neutral gateway | approved-production |
| Low-volume inference Cloudflare Workers AI recurringproduction-approved | 10,000 neurons/day; model availability varies and some models require Workers Paid. Commercial use permitted for approved models under current terms. | Low-volume classification, summarization, metadata, and query rewriting. | Model-specific neuron consumption, resetting daily at 00:00 UTC. Model registry and policy checks are mandatory. | Deterministic output or retrieval-only result. | Neuron use reaches 75%, selected model moves to paid, or quality requirements exceed available models. Likely paid successor: Workers Paid or approved generation adapter | approved-limited |
| Semantic retrieval Cloudflare Vectorize recurringproduction-approved | 30M queried vector dimensions/month and 5M stored vector dimensions. Commercial use permitted. | A small public-content index supporting prototype or low-volume semantic retrieval. | Stored vector dimensions and queried vector dimensions across the account. Embeddings may reveal source semantics; classify content before indexing. | Keyword search, D1 metadata filters, and curated navigation. | Dimension use, corpus size, latency, or quality requires a paid vector store. Likely paid successor: Vectorize paid usage or approved vector adapter | approved-production |
| Embedding bootstrap Voyage AI one-timeproduction-approved | First 200M tokens free for several current models; not a monthly renewal. Commercial use subject to account terms. | Initial embedding or reranking bootstrap for an approved content index. | Input tokens processed by eligible models; the account allowance is one-time rather than monthly. Verify training and retention terms before private content use. | Workers AI embeddings or keyword search. | Bootstrap allowance is consumed or model requirements change. Likely paid successor: Voyage AI paid usage or approved embedding adapter | approved-limited |
| Realtime transcription Speechmatics recurringproduction-approved | 20 realtime hours/month and two concurrent sessions. Commercial use permitted subject to current terms. | Occasional live captions or voice input, with limited batch transcription for supplied recordings. | Realtime minutes, batch minutes, and concurrent realtime sessions. Explicit microphone consent; no background recording; verify regional requirements. | Text input, uploaded transcript, or manual caption workflow. | Hours or concurrency approach 75%, or batch processing is required. Likely paid successor: Speechmatics paid plan or approved transcription adapter | approved-limited |
| Browser verification Cloudflare Browser Run recurringproduction-approved | 10 browser minutes/day on Workers Free. Commercial use permitted. | A narrow daily critical-path schedule covering a few important public pages and forms. | Browser session duration shared across Quick Actions and Browser Sessions. Do not capture sensitive pages or secrets in screenshots. | HTTP checks, manual QA checklist, and user-reported issue path. | Critical-path schedule cannot fit within the daily minute budget. Likely paid successor: Cloudflare paid browser usage or approved monitoring provider | approved-limited |
| Beta cited retrieval accelerator Cloudflare AI Search prototypeproduction-approved | Open-beta allowance described as 20K queries/month, 100K files/instance, and 500 crawled pages/day; terms may change with notice. Approved limited while beta terms remain acceptable. | A controlled beta retrieval prototype over an approved website or document corpus. | Monthly queries, files per instance, and pages crawled; Workers AI and AI Gateway usage remains separate. Never make the production promise depend on beta availability. | Worker + D1/R2 + Vectorize + embedding adapter + conventional search. | Beta terms, pricing, availability, or quality no longer meet requirements. Likely paid successor: Provider-independent retrieval architecture | approved-limited |
Verification dates describe the repository record, not a permanent guarantee. Re-verify every provider before proposal and quarterly during Care. Stale records must be quarantined rather than displayed as current.