A GitHub repository gained 915 stars today by maintaining information that can expire between two visits to a pricing page. free-for-dev is a 253KB Markdown directory of hosted services with free tiers, and one of its latest edits changed a chat tool's monthly AI allowance from 10 replies to 25. The spike is less about a new release than the appeal of having somebody track terms that developers otherwise rediscover service by service.
That maintenance problem gives ripienaar/free-for-dev a sharper purpose than its name suggests. The project does not collect every product with a zero-dollar button. It defines what counts as a usable free tier, rejects trials, and requires time-limited allowances to last at least a year. Its rules also exclude self-hosted software because the list is specifically about services.
A strict definition of free
The current README spans 1,701 lines and sorts services into 57 sections, including cloud providers, source-code hosting, security, monitoring, testing, and generative AI. A count of its linked service bullets yields more than 1,200 entries. The README credits contributions from more than 1,600 people, which helps explain how a single document can cover products ranging from DNS tools to managed databases.
Admission starts with a distinction that pricing pages often blur. A service must have a free tier rather than a free trial. If the free allocation is measured over time, it must remain available for at least one year. The project accepts single sign-on as a paid feature, but its stated security rule rejects services that reserve TLS for paying customers.
The pull-request checklist adds more conditions. Pricing must be visible without signing up or making a phone call, and a submission must say exactly what is free. The service needs operator contact details and a privacy policy. Contributors must also check that the product is absent from the list, according to the project's submission template.
Some exclusions are editorial rather than technical. The maintainers decline generic cPanel-style hosting, temporary email generators, Cloudflare front ends that add no distinct service, and general developer toolbox sites. The contribution guide says the list is aimed mainly at infrastructure developers and that adding another converter or calculator would bring little value. This keeps the directory from becoming a catalogue of anything marketed to programmers.
Those filters matter because the word "free" covers very different deals. One listing may offer a permanent monthly quota, another may require a credit card, and a third may waive fees only for open-source projects. The directory records such conditions inside each description. It is still a discovery tool, not a contract: providers control their current terms, and the repository tells contributors to remove offerings that have changed or retired.
Maintenance is the product
GitHub shows that the repository was created in March 2015 and now has roughly 133,700 stars and 14,000 forks. It has accumulated more than 7,200 commits without publishing conventional GitHub releases. That history fits the format. There is no binary to ship; the useful state is the latest reviewed version of the list and the record of how it changed.
The recent commit log makes that work visible. On August 22, maintainers merged a cloud-cost analyzer and an update to GhostChat's AI allowance. The previous day brought a monitoring service. Other August changes added entries, corrected descriptions, or updated free-plan details. Each edit is small, while stale limits across hundreds of services would compound quickly.
A plain README has benefits here. GitHub can review every alteration as a diff, contributors can propose one-line corrections, and readers can inspect blame or history without learning a custom interface. The same choice has costs. Entries in the current file do not carry a uniform last-verified date, pricing fields are prose, and limits such as requests per month cannot be queried as structured data. A recent commit proves that a line changed; it does not independently verify a provider's pricing claim.
That is why our review of free-for-dev covers the setup and catalogue reality rather than treating the star count as a quality score. The 915 stars recorded today show fresh attention to the project. They do not measure accuracy, coverage, or how quickly maintainers remove expired offers. Developers should follow each listing to the provider before choosing a dependency or estimating operating cost.
The directory's rules reduce one common source of confusion: a promotional trial should not appear beside a standing free plan. They cannot eliminate plan churn. A provider can lower quotas, add card verification, restrict a region, or close a product after an entry is merged. Git history makes corrections transparent, but the repository has no published guarantee that every one of its more than 1,200 linked entries was rechecked within a fixed interval.
An unusually direct ban on AI submissions
The project has drawn renewed attention during an era of automated pull requests, yet its maintainers explicitly reject AI-written contributions. The contribution guide says AI-generated edits will be closed without discussion. It applies the same rule to new submissions and updates, even when the requested change is a short pricing correction.
The repository also includes an AGENTS.md file addressed to coding agents. It instructs them to tell users that AI edits are not accepted and that contributors must read the pull-request template. This is a rare case where a repository's machine-readable agent guidance tells the machine to stop before producing a patch.
The template reinforces that policy with a checkbox labeled for large language models and other AI to tick. In context, the checkbox acts as a disclosure mechanism: the surrounding comments say LLM-written submissions are not accepted. A human contributor still has to check the substantive requirements about pricing, privacy, duplication, and the scope of the free plan.
The ban reflects the repository's moderation policy, not a finding that humans always submit accurate pricing. Service owners can misstate an offer, and well-meaning contributors can copy an outdated limit. The maintainers' concern is traceable responsibility for an edit in a list where a polished, plausible sentence may still be wrong. Their guide warns that misrepresentation or abusive argument can lead to a block.
There is no described automated detector or test suite for enforcing the AI rule. Reviewers appear to rely on the template, the submitted change, and contributor behavior. That approach may become harder to sustain if a 915-star day produces a wave of promotional pull requests. The repository currently keeps its criteria short enough for a reviewer to apply line by line.
How to use the list without outsourcing judgment
The most efficient path is to begin with the relevant README section and treat its quota descriptions as leads. A developer comparing hosted logging services, for example, can use the directory to identify candidates and see whether each free plan is measured by events, retention, seats, or storage. The provider's current pricing and acceptable-use terms should settle the decision, especially when an application could cross the free threshold suddenly.
Repository history can provide a second check. If an entry's wording looks unusually generous, blame shows when that line entered the list and links to the change that proposed it. The pull request may contain discussion or a pricing link. None of that replaces live verification, but it reveals whether the description has been maintained recently or has survived untouched for years.
Teams should also separate evaluation cost from production cost. A free tier may be enough to prototype an API while remaining unsuitable for a deployed service because of data retention, support, regional availability, or rate limits. Free-for-dev's own scope says the services should help developers evaluate and build something users can later use and receive support for. That standard is more demanding than a one-hour demo, though it still leaves the production decision to the reader.
The next thing to watch is how the maintainers absorb the attention. Fresh stars are cheap; reviewing new claims against visible pricing, privacy policies, and durable quotas takes time. Useful progress would show up as prompt corrections to changed plans and a manageable review queue, while a flood of vendor submissions could make the directory less reliable. The commit log, more than the star counter, will show which way it goes.