Abandoned PostgreSQL projects

Plume creates a branched copy of your Postgres database per preview deployment. Like Git branches but for data. Used in 4 production setups I know of. I built it because Neon's branching was expensive and Supabase didn't have it. The interesting engineering was the copy-on-write at the table level. I'm leaving because the company I built this for is shutting down and the use case at my new job doesn't need it.

Built Lumen over 14 months while freelancing. The idea was: drop a schema, get a production GraphQL API with auth, subscriptions, row-level security — no glue code. Got to ~600 stars, a few production users, and an OKR of "hit 1k stars then find a co-maintainer." I never hit the OKR. I took a full-time role in November, and after two months of trying to maintain it on nights and weekends, I let it go. The codebase is solid, the docs are good, the test suite is honest. Someone who wanted to go full indie on it would have a real head start.

Pollen is the CMS I wanted to use for my own freelance work. WYSIWYG-friendly for clients, but stores structured data. Built the editor as a contenteditable wrapper that emits JSON Schema. Adopted by Pollen Labs (I'm still on the board but no longer the maintainer).

Cobalt is a Figma-lite built on Yjs. I built it to teach myself CRDTs. The interesting part was the performance budget — we hit 60fps on 5k objects per user, 8 concurrent users, on a M1. The codebase is small (~8k lines of TS) and the test suite is the best I've ever written. I'm not using it, my co-author is at a startup that acquired it, and we're looking for someone to take it from here.

Vessel schedules containerized tasks with cron syntax and DAG dependencies. Used to power my side project's data pipeline. I built it because AWS Batch was overkill and Airflow was too much. I'm leaving because I don't run side projects with data pipelines anymore. The scheduler works, the queue is solid, the UI is functional. The Docker socket handling is the part I'm most proud of.

Mosaic outputs semantic HTML, not a div soup. The constraint engine enforces grid + flex rules. The components are HTML-first with optional JS sprinkles. I built it for a client project and open-sourced it after. Adoption was low because the docs are bad. The core is good. Needs someone who likes writing component libraries.

Tessera serves MBTiles files as vector tiles with custom Mapbox GL styles. Used it to build a hiking maps site for a national park. The interesting bit was the style compiler that takes a JSON style spec and emits optimized PMTiles. I'm leaving the maintainer role because the OSM ecosystem has moved on to PMTiles and the project would need a rewrite to stay current.

Pinnacle is what I wanted Shopify to be when I was running a tiny t-shirt shop. Clean API, no theme lock-in, supports weird product types. Got to 40 paying customers, then realized I'd built myself a second job. Sold the customer list to a friend with a Shopify store (with their consent) and walked away. The code is well-organized. The docs are okay. The pricing model is wrong — I made it too cheap.

Saffron was my side project for 3 years. I built it because I wanted to send a weekly newsletter to my readers without anyone tracking opens. Pure SMTP, no analytics, no SaaS subscription. I was getting ready to shut it down when Marco from the community asked to take over. He's been shipping releases every two weeks since. The roadmap is now his.

Compass counts page views and unique visitors with zero cookies and zero fingerprinting. I built it because I was tired of Plausible's pricing for a hobby site. Compass was the lowest-effort tool I've ever maintained. A community member (an actual privacy researcher) took it over when I moved on. They've added geo-IP aggregation without compromising the privacy model.