I finally shipped it 🚀 Externa: an open source, Laravel-native headless CMS.
Code-first schemas, React admin, OpenAPI, live preview, scheduled publishing, and an AI assistant that asks before touching anything important.
Feedback from fellow devs very welcome 👇
@dantegaleazzi@bot It’s garbage, you pay more for having less usage since it’s all mixed in one weekly pool. This destroyed the usage of Cursor which is a tool for working
@MikeLopez Yes. ?include= only expands relations into collections the key can read; anything outside its scope stays a bare id, same as a role without read access. GraphQL items only carry ids, so there's no nested path to walk. Did you hit that gap in another CMS?
Externa 1.3.0 is out, the biggest release since launch.
- Scoped API keys: limit a key to specific collections or make it read-only, across REST and GraphQL
- Optional caps for GraphQL depth and complexity, request cost and payload size
- ?include= relations load in one batch, so the query count stays flat as lists grow
- Project settings split into proper sections
- A table-driven authorization test suite in Pest
Release notes and upgrade guide: https://t.co/0zfQxAzN9L
Externa is now a reference corpus for AI coding agents on the XERJ Corpus Hub @xerj_org . When an agent hits a problem like scheduled publish/unpublish in Laravel, it can pull Externa's real implementation, cited to file and line, instead of guessing from memory.
https://t.co/dyxnwVbyM7
https://t.co/s80QUeC7o9
@taylorotwell Worth the morning. What paid off most for me in AGENTS.md: the exact commands that prove a task is done (tests, lint, PHPStan) and a short "never do" list. Less prose, more checks the agent can actually run.
@cursor_ai@xai Auto quality also dropped sharply: code that ignores rules set in Cursor, where before Auto was great 99% of the time. Please give devs a dedicated Cursor allowance, or a coding-only plan. I really don't want to move to Claude Code.
@cursor_ai@xai I pay for X Premium and https://t.co/BJeIQGObK1 features I don't use, while the tool I need for work got less usable capacity. And the switch is irreversible. "More usage" on paper, much less in practice.
@enunomaduro@AmpCode Curious what clicked for you compared to Claude Code or Cursor: the shared threads, or how it handles long-running tasks? Looking forward to the video.
@PovilasKorop Thanks for sharing, Povilas! Quick context for anyone curious: Laravel-native, React admin, REST + GraphQL, live preview and scheduled publishing. 1.3.0 just landed with scoped API keys. Happy to answer questions here.
@mattiasgeniar@dhh +1. This week showed how much of the web rests on a few PHP internals folks patching overnight. If your company bills on PHP, a Foundation sponsorship is cheaper than one incident.
@taylorotwell Cost pushes toward routing more than raw volume: cheap or local models for the mechanical work, frontier models where reasoning matters. The winners move more useful tokens per dollar, not just more tokens
@marcelpociot Agree the slop fear is overstated. What keeps it maintainable for me are guardrails the agent can't skip: arch tests for naming and layering, PHPStan, a test that fails before the fix. The agent writes more code, the rules decide what stays.
@PovilasKorop The part nobody prices in: review time doesn't scale with generation. My rule now: if I can't review it in the time it took to spec it, the task was too big.