Scheduled jobs, data pipelines and monitors — work that happens on a timer instead of in a queue. Fifteen of them run my own operation right now: pulling analytics, publishing content, watching competitors, backing up databases, and shutting down idle containers to reclaim GPU memory.
Everything on this page runs my own business — it is not client work. It is the same stack, built the same way, that I would set up for you. The numbers below are live row counts from the databases these jobs write to, not estimates.
Grouped by what they actually do. Every one of them last ran successfully.
Runs daily. Picks a keyword, checks what already exists, generates a 2,500+ word article with schema and images, and pushes it to WordPress as a draft for review. The publish decision stays human.
Daily 08:00 · keyword to draftCloudflare traffic and Google Search Console data pulled automatically and written to a database, so reporting reads from stored history instead of a manual export each week.
Daily 12:00 · traffic + searchWatches the third-party sources a site depends on for facts, snapshots them, and diffs each run. When a source changes, the pages that quote it get flagged instead of silently going stale.
Mondays 12:00 · snapshot and diffBuyer-intent posts on Reddit every 12 hours, competitor product movement on the marketplaces weekly, and new job postings every three days — delivered as alerts, not dashboards nobody opens.
12h / weekly · alerts onlyDatabase dumps daily, application state backed up nightly, off the machine to separate storage. A backup you have not restored is a guess, so these report failures loudly instead of failing quietly.
Daily 03:00 · verified dumpsChecks a fleet of GPU and model containers every 15 minutes, parks the ones that have been idle, and wakes them on demand. Idle containers holding VRAM is the difference between a render queue and a queue that stalls.
Every 15 min · resource reclaimWeekly sweep of the automation estate itself — dead webhooks, orphaned workflows, drifted configuration. The maintenance job that stops the maintenance jobs becoming the problem.
Weekly · self-auditA statistical sports model runs weekly, produces forecasts against stored odds, and settles itself against real results the following week — so the track record is kept by the system, not by whoever remembers.
Weekly · forecast and settleA weekly performance report assembled and delivered automatically, plus a sweep of the automation estate for bloat — dead jobs, oversized configs, things that quietly stopped being worth their runtime.
Mondays 14:00 · report deliveredThese jobs are run by Hermes, an open-source AI agent framework — it plans the work, calls the tools, writes the results to a database, and tells me in plain language when something stops. n8n moves data between the services, PostgreSQL is the memory, and the models run on a GPU in the room.
A cron job fires and forgets. The agent reads what came back and does something with it — retries what failed, flags what looks wrong, skips what hasn't changed instead of redoing it.
Every run leaves a row: what it did, what it found, what it cost. That is why I can say what broke three weeks ago — and why the same mistake doesn't come back twice.
Anything that publishes or spends runs with a review step or a dry run first. An agent left unguarded is a liability, not a feature.
Enquiries from a form, inbox or chat landing in one place, enriched, de-duplicated and routed — with an alert when something sits unanswered.
Repurposing pipelines, scheduled publishing, and asset generation that turns one long recording into a week of posts without a human in the loop.
Traffic, search, revenue and ad data pulled on a schedule into one store, with a report that arrives instead of a spreadsheet you maintain.
Uptime, price and competitor watching, source-change detection, backup verification — the jobs that tell you something broke before a customer does.
Hundreds of clips or images resized, converted, captioned or upscaled on a local GPU — no per-file API bill, no upload queue.
Wiring systems that do not talk to each other: databases, CRMs, webhooks, spreadsheets, WordPress. Reported honestly when the API cannot do what was asked.
Honest limits. I automate around APIs, and an API that refuses a task is a limit, not a challenge. No scraping of sites that block it, no guarantee a third-party endpoint stays stable, and no unattended automation on a process that needs a judgement call — publishing, pricing and outbound messages all keep a human gate. I will tell you which parts of your process can actually be automated before quoting, because some of them cannot.