LinkedIn Posts Automation

The obvious way to post on a schedule is to drive a browser and click the buttons for somebody.

This does not do that.

Posts go out through LinkedIn's official Posts API, so the account is doing something the platform has a documented route for rather than something that only looks like a person typing. Nothing here scrapes a page. Nothing here signs in as you.

That decision costs the project every convenience a scraper would have bought, and it is the reason the rest of the design looks the way it does.

A Cloudflare Worker holds the schedule.

It builds a daily plan weighted by engagement, inside windows you set yourself, then publishes what the plan picked: the post, its card image, and the link placed in the first comment rather than in the body.

Why the first comment helps is not a claim I will make here. It is what the tool does. What LinkedIn does with it afterwards belongs to LinkedIn.

Supabase is the source of truth for selection.

That sounds like plumbing until you think about what a scheduler is for. A post published twice is worse than a post never published, because the second copy is visible to everybody who saw the first and there is no undo that removes the memory of it.

So selection happens against the database, and a published item is never picked again.

The words themselves are written in advance. A pipeline reads pre-written post files, seeds the database from them, and the scheduler works through what is there. Nothing generates text at publish time.

That is a smaller product than one that writes for you, and smaller on purpose.

A queue of sentences somebody meant is worth more than a queue a machine filled overnight, and the scheduler's only job is deciding when each of them goes out.

The dashboard is a real application rather than a settings file.

It is React with Capacitor, signed in through Firebase, reading over TanStack Query, and it exists so cadence can be watched, priorities and windows edited, and a single post skipped or pushed back into the queue without anyone going near the worker.

There is a second way in for the times a dashboard is the slowest thing to reach for: an operator API behind a bearer token, giving status, logs, the current plan and a manual run. Configurable email notifications carry the digests, the finished runs and the warnings.

Now the part that does not go away.

LinkedIn's authorisation is renewed by hand, roughly every 60 days. The system warns before the token expires and halts cleanly on a 401 instead of hammering a dead credential.

It cannot renew itself.

This is automation with a standing appointment, which is the honest description and the thing a page like this usually leaves out, because runs itself reads better than runs itself until the sixty-first day.

Writing that limit down was cheaper than being surprised by it twice. The documentation says so in the same place it says everything else the API will not do.

The whole stack sits on free tiers. A Worker on a cron, a Postgres database, a small dashboard: none of it needs a server kept alive, and none of it carries a bill that grows with the number of posts.

It is at a1la.aoneahsan.com.

The re-authorisation is still in the calendar, and it always will be.