One Dev Runs Bluesky's For You Feed From a Living Room PC for $30/Month

A developer known as spacecowboy runs one of Bluesky's most popular features — the "For You" recommendation feed serving 72,000 daily active users — from a single desktop PC in their living room.

The total cost? About $30 per month. Equivalent cloud infrastructure would run roughly $245.

The algorithm is elegantly simple: it finds people who liked the same posts as you, then shows you what else they've liked recently. No complex AI models, no GPU clusters.

Under the hood, a single Go binary handles everything — consuming Bluesky's firehose, storing data in a 419GB SQLite database, and serving recommendations. In-memory LRU caches handle 99% of queries, keeping the system snappy at 15-25 requests per second while using only 37% CPU.

The hardware is a consumer AMD 9950X3D with 96GB RAM. A battery backup provides 4-5 hours of runtime during power outages. A small OVH VPS acts as a reverse proxy to keep the home IP hidden.

The result: 99.77% uptime since January 2026, with theoretical headroom to scale to 3x current traffic — or even 30x with an adaptive algorithm that reduces computational costs by over 10x.

This is possible because Bluesky's AT Protocol lets anyone build and publish custom feed algorithms. It's a fundamentally different model from centralized platforms where only the company controls what you see.

spacecowboy keeps it as a pure hobby project with no monetization — proving that meaningful infrastructure doesn't always require massive investment.