--- title: "Free compute worth claiming in August 2026" date: 2026-08-03 description: "An August 2026 field guide to free VPS, serverless, databases, agent sandboxes, developer tools, AI APIs, and GPU notebooks." tags: - "self-hosting" - "free-tier" - "side-projects" - "cloud-infrastructure" - "ai-tools" --- This is the 3 August 2026 snapshot of free infrastructure worth claiming for side projects. It is the third post in the free-compute thread, following the [June guide](/posts/free-compute-for-ai-side-projects/) and the [July guide](/posts/free-compute-july-2026/), and I wanted to do it again for August while the July version was still fresh. That turned out to matter. GitHub Models is gone. SageMaker Studio Lab has closed to new users. The current Hugging Face documentation no longer supports the broad free CPU Spaces claim I made last month. On the other side, Paperspace has a properly documented free M4000 notebook, Neon has opened Functions and Object Storage as a free beta, and agent and browser compute keeps becoming a category of its own. I also need to correct the July post. GitHub announced on 1 July that Models would retire on 30 July, but the 10 July guide still described it as a recurring public preview. That was wrong. The service is now fully retired, including the playground, catalogue, inference API, and BYOK endpoints. The changes and retained offers below were checked against official provider pages on the snapshot date. I have not opened and tested every account. Regions, capacity, billing checks, data-use terms, and unpublished guardrails can still change what is available to you. ## What changed for August - [GitHub Models retired on 30 July](https://github.blog/changelog/2026-07-01-github-models-is-being-fully-retired-on-july-30-2026/). It no longer belongs in a free model API list. - [SageMaker Studio Lab](https://docs.aws.amazon.com/sagemaker/latest/dg/studio-lab-overview.html) also closed to new customers on 30 July. Existing users keep their free CPU and GPU sessions, but there is nothing left for a new reader to claim. - [Hugging Face Spaces](https://huggingface.co/docs/hub/spaces-overview) needs narrowing. The current documentation gates normal Gradio and Docker compute Spaces behind paid plans. Free personal accounts can still host up to two [Gradio ZeroGPU Spaces](https://huggingface.co/docs/hub/main/en/spaces-zerogpu), but that is shared demo GPU capacity rather than a free Linux box. - [Paperspace Notebooks](https://docs.digitalocean.com/products/paperspace/pricing/) is a new catch. Its Free Gradient plan makes C4 CPU and M4000 GPU machines available, includes 5 GB storage, and allows [one notebook to run at a time](https://docs.digitalocean.com/products/paperspace/notebooks/getting-started/run-example-notebooks/). [Free machines can run for up to six hours](https://docs.digitalocean.com/products/paperspace/notebooks/details/features/), but capacity may queue and there is no published monthly-hour guarantee. - [Neon Functions and Object Storage](https://neon.com/blog/neon-backend-is-beta) are now available on the Free plan during beta. They are limited to `us-east-2`, keep logs for three days, and have unpublished rate and usage guardrails. AI Gateway is not included on Free, despite the broad full-backend launch language. - [Google Agent Engine](https://docs.cloud.google.com/free/docs/free-cloud-features) is listed in Google's Free Tier table with 180,000 vCPU-seconds and 360,000 GiB-seconds per month. Billing must be enabled. The table does not prove that Code Execution sandboxes share the same allowance, so I am treating this narrowly as managed agent runtime. - [Browserbase](https://www.browserbase.com/pricing) and [Browserless](https://cloud.browserless.io/pricing) join Cloudflare Browser Run as useful browser compute. Browserbase gives one browser hour and three agent runs per month. Browserless uses a 1,000-unit pool with short sessions rather than promising a neat number of hours. - [Azure Pipelines](https://learn.microsoft.com/en-us/azure/devops/pipelines/licensing/concurrent-jobs?view=azure-devops) no longer offers the same public-project path to new users. Existing public projects keep their grants only until conversion to private in 2027. Private hosted projects can still get 1,800 minutes per month after Azure billing is linked. - [GitLab](https://about.gitlab.com/blog/efficient-free-tier/) starts enforcing its five-user limit for private Free top-level namespaces on 15 August. Public namespaces are exempt. ## The short version - Claim [OCI Always Free](https://docs.oracle.com/en-us/iaas/Content/FreeTier/freetier_topic-Always_Free_Resources.htm) if you can get capacity and want a real Linux VM. The current supported figure is 2 Ampere OCPUs and 12 GB RAM, not the older 4 OCPU and 24 GB claim. - Use [Cloudflare Workers](https://developers.cloudflare.com/workers/platform/limits/) for small APIs and glue, [Deno Deploy](https://deno.com/deploy/pricing) when its TypeScript-first platform fits, or [Google Cloud Run](https://cloud.google.com/run/pricing) when you need a container and accept a billing-linked allowance. - Use [Neon](https://neon.com/pricing) for intermittent Postgres and try its free beta Functions and Object Storage if `us-east-2` and unpublished limits are acceptable. Use [Turso](https://turso.tech/pricing) when distributed SQLite is the better fit. - Use [Modal](https://modal.com/pricing) for $30 of recurring serverless CPU or GPU credit. Use Paperspace when a notebook with an M4000 is enough and dynamic availability is acceptable. - For disposable agent machines, look at [Upstash Box](https://upstash.com/pricing/box), [Vercel Sandbox](https://vercel.com/pricing), and [Freestyle VMs](https://www.freestyle.sh/pricing). For browser work, start with [Cloudflare Browser Run](https://developers.cloudflare.com/browser-run/pricing/), Browserbase, or Browserless. - Put hard-capped services around anything attached to billing. Free usage inside AWS, Azure, Google Cloud, or CockroachDB is still an account capable of producing a bill. ## Persistent VMs and cloud credits - [Oracle Cloud Infrastructure](https://docs.oracle.com/en-us/iaas/Content/FreeTier/freetier_topic-Always_Free_Resources.htm): **Always Free.** Up to two AMD micro VMs, plus 1,500 Ampere A1 OCPU-hours and 9,000 GB-hours per month, equivalent to 2 OCPUs and 12 GB RAM. The tenancy also gets 200 GB combined boot and block storage and 10 TB monthly outbound transfer. Compute must live in the home region, capacity can be unavailable, and idle instances can be reclaimed. - [Google Compute Engine](https://docs.cloud.google.com/free/docs/free-cloud-features): **Billing-linked recurring allowance.** One non-preemptible `e2-micro` instance-month, 30 GB-month standard persistent disk, and 1 GB outbound per month in selected US regions. Billing is required and usage beyond the allowance can charge. - [AWS EC2 T4g](https://aws.amazon.com/ec2/instance-types/t4/): **Promotion through 31 December 2026.** The offer covers 750 aggregate `t4g.small` hours per month. EBS, public IPv4, non-covered images, and surplus CPU in Unlimited mode can still cost money. This is useful now and not durable. - [AWS Free plan](https://docs.aws.amazon.com/awsaccountbilling/latest/aboutv2/free-tier-plans.html), [Azure free account](https://azure.microsoft.com/en-us/free/), and the [Google Cloud trial](https://docs.cloud.google.com/free/docs/free-cloud-features): **New-account credits.** These are broad evaluation budgets, not permanent free VPSes. AWS closes or upgrades the account after six months or when its credit is exhausted; Azure's $200 lasts 30 days; Google's $300 lasts 90 days. - [DigitalOcean](https://www.digitalocean.com/pricing), [Akamai Cloud](https://www.akamai.com/lp/free-credit-100), [OVHcloud](https://www.ovhcloud.com/en-gb/public-cloud/free-trial/), [Vultr](https://docs.vultr.com/support/platform/billing/how-can-i-add-a-promotional-code), and [Fly.io](https://fly.io/docs/about/free-trial/): **Trials and promotions.** Use them for provider evaluation or a short burst. None is a current permanent free VPS offer for a new account. ## Functions, containers, and PaaS - [Cloudflare Workers](https://developers.cloudflare.com/workers/platform/limits/): **Recurring hard cap.** 100,000 requests per day, 10 ms CPU per HTTP request, and 128 MB memory. Static assets do not consume the dynamic request allowance. [Workflows](https://developers.cloudflare.com/workflows/platform/pricing/) separately gives 3,000 steps per day and 1 GB state storage on Workers Free. - [Deno Deploy](https://deno.com/deploy/pricing): **Recurring hard cap.** 1 million requests, 20 GB egress, 15 CPU-hours, 350 GB-hours memory, 20 apps, 1 GiB volume storage, and 1 GiB KV storage per month. Deploy Classic's 20 July migration deadline has passed and old projects did not transfer automatically. - [Google Cloud Run](https://cloud.google.com/run/pricing): **Billing-linked recurring allowance.** Request-based services get 2 million requests, 180,000 vCPU-seconds, and 360,000 GiB-seconds per month. Build, registry, storage, and network usage can still create charges. - [Azure Functions](https://azure.microsoft.com/en-us/pricing/details/functions/) and [Azure Container Apps](https://azure.microsoft.com/en-us/pricing/details/container-apps/): **Billing-linked recurring allowances.** Classic Functions includes 1 million executions and 400,000 GB-seconds. Container Apps includes 2 million requests, 180,000 vCPU-seconds, and 360,000 GiB-seconds. Storage, networking, and minimum replicas sit outside those grants. - [IBM Cloud Code Engine](https://cloud.ibm.com/docs/codeengine?topic=codeengine-pricing): **Billing-linked recurring allowance.** 100,000 vCPU-seconds, 200,000 GB-seconds memory, and 100,000 HTTP requests monthly across apps, jobs, functions, and builds. - [Scaleway Functions, Containers, and Jobs](https://www.scaleway.com/en/pricing/serverless/): **Billing-linked recurring allowances.** Functions gets 1 million requests and 400,000 GB-seconds. Containers and Jobs each add CPU and memory grants, but warm or provisioned instances charge continuously. - [Vercel Hobby](https://vercel.com/pricing): **Personal and non-commercial hard caps.** It remains a strong frontend and function platform for personal experiments, but its terms make it a poor place to hide a commercial product. [Vercel Sandbox](https://vercel.com/docs/sandbox) has a separate Hobby allowance for disposable microVMs. - [Netlify Free](https://docs.netlify.com/manage/accounts-and-billing/billing/billing-for-credit-based-plans/how-credits-work/): **Recurring shared hard cap.** Static delivery, functions, Agent Runners, Preview Servers, and AI inference all consume the same 300 monthly credits. Projects pause when the pool is empty. - [Render](https://render.com/docs/free) and [Railway](https://docs.railway.com/pricing): **Small non-production allowances.** Render's free web services sleep after 15 idle minutes. Railway starts with a one-time $5 trial, then provides a recurring $1 monthly credit and can block free deployments during regional peaks. ## Agent sandboxes and browser compute - [Cloudflare temporary accounts](https://developers.cloudflare.com/workers/platform/claim-deployments/): **One-hour preview.** An unauthenticated agent can run `wrangler deploy --temporary`, deploy a Worker and supported resources, verify them, and return a claim URL. Unclaimed accounts are deleted after 60 minutes. The preview can include KV, one 100 MB D1 database, Durable Objects, Hyperdrive, and Queues. This is throwaway deployment compute, not a permanent anonymous account. - [Upstash Box](https://upstash.com/pricing/box): **Free developer preview.** Ten concurrent boxes and five active CPU-hours per month. Each box has 2 vCPU, 4 GB RAM, and 5 GB disk, freezes after one idle hour, and currently runs only in `us-east-1`. - [Vercel Sandbox](https://vercel.com/pricing) and [Freestyle VMs](https://www.freestyle.sh/pricing): **Recurring disposable compute.** Vercel's Hobby allowance is for personal, non-commercial use. Freestyle publishes daily CPU, memory, and storage allowances, but free VMs and snapshots are not persistent. - [Cloudflare Browser Run](https://developers.cloudflare.com/browser-run/pricing/): **Recurring hard cap.** Ten browser minutes per day and three concurrent browsers. Cloudflare's general-purpose Sandbox SDK is different and currently has no Workers Free allocation. - [Browserbase](https://www.browserbase.com/pricing): **Recurring hard cap.** One browser hour, three concurrent browsers, three agent runs, and 1,000 each of Search and Fetch uses per month. Sessions stop at 15 minutes. - [Browserless](https://cloud.browserless.io/pricing): **Recurring unit pool.** 1,000 units per month, two concurrent browsers, and one-minute sessions. A unit covers up to 30 seconds, while proxy and CAPTCHA work can consume more, so the allowance does not translate cleanly into hours. - [E2B](https://e2b.dev/pricing) and [Daytona](https://www.daytona.io/pricing): **One-time or unclear credits.** E2B's $100 is explicitly a starter credit. Daytona advertises $200 but still does not document recurrence or expiry clearly. Neither belongs in the recurring-free group. ## Databases, queues, and storage - [Neon](https://neon.com/pricing): **Recurring, no card.** Up to 100 projects, each with 100 CU-hours, 0.5 GB storage, compute up to 2 CU and 8 GB RAM, and five-minute scale-to-zero. An always-on database would exhaust that compute allowance. The new Functions and Object Storage beta is free for now, but has no published quota or GA price. - [Aiven PostgreSQL and MySQL](https://aiven.io/docs/platform/concepts/service-pricing): **Recurring, no card.** One free service of each type per organization, each with 1 CPU, 1 GB RAM, 1 GB storage, and backups. Free services can power off when idle and have no SLA. - [Turso](https://turso.tech/pricing): **Recurring, no card.** 100 databases, 5 GB storage, 500 million rows read, 10 million rows written, and 3 GB embedded sync per month. Requests stop at the quota instead of rolling into an invoice. - [Cloudflare D1](https://developers.cloudflare.com/d1/platform/pricing/): **Recurring hard cap.** Ten databases, 500 MB each, 5 GB total, 5 million rows read and 100,000 rows written per day. Use it when the rest of the app is already on Workers. - [Upstash](https://upstash.com/pricing): **Recurring product plumbing.** The free bundle covers Redis, QStash, Workflow, Vector, and preview Search with separate quotas. Free Redis databases can be archived after inactivity, and Search has no SLA while in preview. - [Qdrant Cloud](https://qdrant.tech/documentation/cloud/create-cluster/) and [Pinecone Starter](https://docs.pinecone.io/reference/api/database-limits): **Recurring vector stores.** Qdrant's free 1 GB RAM, 0.5 vCPU, 4 GB disk cluster suspends after an idle week and can be deleted after four idle weeks. Pinecone gives five indexes and 2 GB storage in AWS `us-east-1`, with monthly read, write, and embedding allowances. - [Cloudflare R2](https://developers.cloudflare.com/r2/pricing/): **Billing-linked recurring allowance.** 10 GB-month Standard storage, 1 million Class A operations, 10 million Class B operations, and free egress. Enabling it requires account checkout even if usage stays inside the allowance. - [Tigris](https://www.tigrisdata.com/pricing/) and [Backblaze B2](https://www.backblaze.com/cloud-storage/pricing): **Recurring object storage.** Tigris provides 5 GB with small operation allowances and free egress. B2 provides the first 10 GB and free monthly egress up to three times average stored data. - [Cloudflare DNS](https://developers.cloudflare.com/dns/faq/): **Recurring.** Authoritative DNS queries remain uncapped and unbilled on normal plans, alongside CDN, TLS, and DDoS protection. Free-zone record limits depend on when the zone was created. ## Email, auth, observability, and CI - [Resend](https://resend.com/docs/knowledge-base/what-is-resend-pricing), [Brevo](https://help.brevo.com/hc/en-us/articles/208580669-FAQs-What-are-the-limits-of-the-Free-plan), and [Mailgun](https://help.mailgun.com/hc/en-us/articles/203068914-What-does-the-Free-plan-offer): **Recurring email.** Resend gives 3,000 sent and received emails per month with a 100/day limit. Brevo gives 300 sends/day with branding. Mailgun gives 100 messages/day and one custom domain. - [Clerk](https://clerk.com/pricing), [Firebase Authentication](https://firebase.google.com/pricing), and [WorkOS AuthKit](https://workos.com/pricing): **Recurring auth.** Clerk includes 50,000 monthly retained users per app. Firebase Spark includes 50,000 MAU for non-phone auth. AuthKit includes 1 million MAU, although production requires billing information. - [Svix](https://www.svix.com/pricing/) and [Hookdeck](https://hookdeck.com/pricing): **Recurring webhook plumbing.** Svix includes 50,000 attempted outgoing messages and 30-day payload retention. Hookdeck includes 10,000 incoming events and three-day retention. - [PostHog](https://posthog.com/) and [Flagsmith](https://www.flagsmith.com/pricing): **Recurring product telemetry.** PostHog includes 1 million analytics events, 5,000 replays, and 1 million feature-flag requests. Flagsmith includes 50,000 flag requests, one member, and one project; Free accounts get seven days after an over-limit notice before requests are blocked. - [Grafana Cloud](https://grafana.com/pricing/), [Better Stack](https://betterstack.com/pricing), and [Sentry](https://sentry.io/pricing/): **Recurring observability.** Grafana includes 10,000 metric series and 50 GB each of logs, traces, and profiles. Better Stack includes ten uptime monitors and ten heartbeats plus small telemetry allowances. Sentry Developer includes 5,000 errors, 5 GB logs, 5 million spans, and 50 replays for one user. - [Axiom Personal](https://axiom.co/docs/reference/limits), [Honeycomb](https://www.honeycomb.io/pricing), and [New Relic](https://newrelic.com/pricing): **Recurring observability.** Axiom currently includes two datasets, not the three I listed in July, with 500 GB monthly ingest and 25 GB stored in a US-only deployment. Honeycomb includes 20 million events. New Relic includes 100 GB ingest and one full-platform user. - [GitHub Actions](https://docs.github.com/en/billing/concepts/product-billing/github-actions) and [Codespaces](https://docs.github.com/en/billing/reference/product-usage-included): **Recurring developer compute.** Standard hosted runners are free for public repositories; private GitHub Free repositories get 2,000 minutes. Personal GitHub Free accounts get 120 Codespaces core-hours and 15 GB-month storage. - [GitLab CI](https://docs.gitlab.com/ci/pipelines/compute_minutes/), [CircleCI](https://circleci.com/pricing/), and [Google Cloud Build](https://cloud.google.com/build/pricing): **Recurring CI.** GitLab gives 400 compute minutes, with the August private-namespace user-limit change noted above. CircleCI gives 30,000 credits, worth roughly 6,000 small Docker minutes. Cloud Build gives 2,500 `e2-standard-2` minutes after billing is linked. - [Azure Pipelines](https://learn.microsoft.com/en-us/azure/devops/pipelines/licensing/concurrent-jobs?view=azure-devops): **Billing-linked or legacy-only.** Private projects can get 1,800 hosted minutes after linking an Azure subscription, plus one free self-hosted parallel job. The uncapped public-project grant applies only to existing projects and ends when they convert to private in 2027. ## AI, GPU, and model APIs Free AI compute is still several different products wearing one label. A notebook lends you a machine. ZeroGPU lends a public demo a queued slice of hardware. Modal gives a recurring dollar credit. Model APIs provide inference without exposing the computer at all. ### GPU and notebook compute - [Modal Starter](https://modal.com/pricing): **Recurring $30 compute credit.** The credit can pay for serverless CPU, GPU, notebooks, schedules, sandboxes, and web functions. Starter includes 100 containers, ten-GPU concurrency, three seats, and one-day log retention. - [Lightning AI Free](https://lightning.ai/pricing/): **Recurring credit.** Fifteen credits per month plus one active CPU Studio that restarts every four hours. The GPU time those credits buy varies sharply by hardware and current rates, so the credit is the stable figure to quote rather than the marketing hour headline. - [Paperspace Notebooks](https://docs.digitalocean.com/products/paperspace/pricing/): **Shared recurring notebook.** Free C4 CPU and M4000 GPU machines, 5 GB storage, one concurrent notebook, and up to six hours per free-machine session. Capacity is not guaranteed and no monthly-hour number is published. - [Google Colab](https://research.google.com/colaboratory/faq.html): **Shared, dynamic notebooks.** Free CPU, GPU, and TPU access with no stable hardware or quota promise. Sessions can run for up to 12 hours depending on usage and availability. - [Databricks Free Edition](https://docs.databricks.com/aws/en/getting-started/free-edition-limitations): **Shared, non-commercial compute.** Limited CPU and capacity-dependent GPU notebooks with unpublished fair-use limits. GPU access where offered requires successful LinkedIn verification. Outbound internet is restricted, and [Free Edition data may be used for training](https://docs.databricks.com/aws/en/getting-started/free-trial-vs-free-edition). - [Hugging Face ZeroGPU](https://huggingface.co/docs/hub/main/en/spaces-zerogpu): **Shared public-demo GPU.** Free personal accounts can host up to two Gradio ZeroGPU Spaces. This is useful for a public demo and not equivalent to an arbitrary GPU notebook or VM. - [SageMaker Studio Lab](https://docs.aws.amazon.com/sagemaker/latest/dg/studio-lab-overview.html): **Existing users only.** Existing accounts keep 15 GB storage, 16 GB RAM, up to eight CPU hours or four GPU hours per day, subject to capacity. New access is closed and AWS plans no new features. ### Hosted model APIs - [Cloudflare Workers AI](https://developers.cloudflare.com/workers-ai/platform/pricing/): **Recurring hard cap.** 10,000 Neurons per day, resetting at midnight UTC. Requests stop when a Workers Free account exhausts the allocation. Cloudflare says it does not train on customer content without consent. - [Gemini Developer API](https://ai.google.dev/gemini-api/docs/pricing): **Model-dependent free tier.** Selected models have free input and output, while limits vary by model, account, and region. Free-tier content may be used to improve Google products. - [GroqCloud](https://console.groq.com/docs/rate-limits) and [OpenRouter Free](https://openrouter.ai/docs/api/reference/limits): **Rate-limited inference.** Groq publishes model-specific organization limits. OpenRouter gives 50 free-model requests per day unless the account has purchased credit. - [FreeInference](https://freeinference.org/): **Experimental research service.** No-card access for open-source, research, and education, with an unpublished quota. Its [terms](https://freeinference.org/terms) allow prompts and responses to be logged and anonymised derived data to be released. Never send it secrets, private code, or personal data. - [Hugging Face Inference Providers](https://huggingface.co/docs/inference-providers/en/pricing): **Tiny recurring credit.** Free accounts receive $0.10 per month, subject to change. It is routed model API credit, not GPU access. - [Cerebras Inference](https://www.cerebras.ai/pricing): **One-time trial.** Current pricing offers $5 trial credit, followed by a minimum $10 deposit for developer usage. Older documentation still describes a free tier, so I am keeping it out of the recurring list. ## Startup, academic, research, and open-source credits These can be much larger than normal free tiers, but they require eligibility, approval, and usually a project or company that already exists. - [AWS Activate](https://aws.amazon.com/startups/credits/) and [Google for Startups Cloud](https://cloud.google.com/startup/ai): application-based cloud credits. The headline maximums are large, but programme tier, affiliation, milestones, expiry, and prior credit use determine the actual award. - [Microsoft for Startups](https://learn.microsoft.com/en-us/startups/microsoft-for-startups/getting-started-mfs): a milestone path rather than an instant $150,000 grant. It begins at $1,000, then unlocks $5,000, $25,000, $50,000, and $150,000 only after verification and sustained Azure workload thresholds. A card is required and the account converts to pay-as-you-go. - [DigitalOcean Hatch](https://www.digitalocean.com/blog/hatch-accelerate-your-ai-ml-startup) and the [OVHcloud Startup Program](https://startup.ovhcloud.com/en/startups/): selected startups can receive much larger time-bounded credits. OVHcloud's current path offers EUR 10,000 for selected pre-seed or seed companies and up to EUR 100,000 for selected later-stage companies, but excludes its A100, H100, L4, and L40S GPU instances. - [Modal academic grants](https://modal.com/academics): up to $10,000 for graduate students, labs, and researchers. [AMD's AI Developer Program](https://developer.amd.com/resources/) separately lets approved members choose either $100 of AMD Developer Cloud credit, expiring 30 days after activation, or a $50 Fireworks code that must be redeemed within 90 days. Once redeemed, the Fireworks credit remains until consumed. - [Google Cloud research credits](https://cloud.google.com/edu/researchers), [AWS Cloud Credit for Research](https://aws.amazon.com/government-education/research-and-technical-computing/cloud-credit-for-research/), [DigitalOcean open-source credits](https://www.digitalocean.com/open-source/credits-for-projects), and [Hugging Face Community GPU Grants](https://huggingface.co/docs/hub/main/spaces-gpus): application-only routes for qualifying research or public projects. Treat an award as runway after approval, not infrastructure available merely because a form exists. ## Stacks that still fit together - **A small real server:** OCI for the VM, Cloudflare for DNS, CDN, and TLS, R2 or Tigris for objects, Neon or Aiven for Postgres, Upstash for queues and Redis, Resend for mail, and Grafana Cloud plus Sentry for visibility. - **A no-server product:** Cloudflare Workers and static assets, D1 or Neon, R2, Queues or Workflows, Turnstile, Clerk or AuthKit, and Workers AI. Temporary Cloudflare accounts now give an agent a one-hour deploy-and-verify loop before the permanent account exists. - **An AI demo:** Paperspace or Colab for notebook work, Modal for repeatable burst jobs, ZeroGPU for a public Gradio surface, Workers AI, Gemini, Groq, or OpenRouter for hosted inference, and R2 plus Neon or Turso for state. - **An agent experiment:** Upstash Box, Vercel Sandbox, or Freestyle for disposable code execution; Browser Run, Browserbase, or Browserless for browser work; Axiom, Honeycomb, or Sentry for the traces that explain what happened. In 24 days, one model API disappeared, one notebook closed to new users, and a supposed free hosting offer became much narrower. August still found a free M4000 notebook, a temporary full-stack beta, managed agent runtime, and more browser compute. The next version may remove as much as it adds. That is fine. A dated guide is more useful when it is willing to get shorter.