I run a solo consulting practice. I have dozens of active clients at any given time. Their data lives across Gmail, Slack, Teamwork, Asana, Google Calendar, meeting transcripts, time tracking, and other platforms depending on the client. I use Trello internally to manage my own work across all of them, but that is too many tabs. I needed one place to see everything.
The constraints
I did not want another SaaS product. I wanted something I could run on the cheapest VPS available and forget about.
The key design decision was making data flow in one direction only. Every source feeds into the pipeline. The pipeline writes to Git repos. A dashboard reads from those repos. Nothing writes back.
Trello is the only write interface. If I need to add a task, mark something done, or reprioritize, I do it in Trello. The pipeline picks up the change within five minutes.
What it actually does
The system runs on a 2GB Linode VPS. Total cost is $14.50 a month including backups.
Cron jobs
Four cron jobs handle everything. They are a mix of bash and Python scripts.
The first pulls raw action items from Trello every five minutes.
The second assembles a cross-client agenda from those items plus Google Calendar on a two-minute offset.
The third runs hourly and does the heavy lifting: importing from every client platform, triaging new items into Trello cards, exporting board state, running Claude Code enrichment, and committing to Git.
The fourth routes calendar events to client queues every fifteen minutes.
Each client gets its own private GitHub repo. The repos are the source of truth. Everything is markdown and structured data with Git as the audit trail.
The dashboard
A Python/Flask script behind nginx and Cloudflare serves as the read-only frontend. It renders markdown, shows the cross-client agenda, and lets me view action items per client.
I can pull it up on my phone or laptop. It is password and IP gated.
Claude Code as infrastructure
The enrichment step in the main sync job runs Claude Code headless on the VPS. It reads the raw data the earlier pipeline steps produced and generates structured summaries, flags priority items, and surfaces things I might miss scanning dozens of boards manually.
This runs unattended. No interactive prompts, no human in the loop during the sync cycle.
What I would do differently
Not much. The one-directional data flow has held up well. Trello as the single write interface means I never have conflicting state between systems. Git as the persistence layer means I have full history and can recover from anything.
If the VPS died tomorrow, I would have it rebuilt in under an hour. Every credential is a file on disk, copied via scp. The VPS itself has automated backups. Everything is also backed up to my AWS S3 and Google Cloud.
Why it works
I own every piece of this. The data, the pipelines, the hosting. No vendor lock-in, no monthly per-seat pricing, no feature deprecations to work around. Just a cheap server doing exactly what I need it to do.
And when I want new features — which I have plenty of in the works — I add them myself on my own timeline.
