Tooling · 2026
Ledger
Problem. Every Go service eventually grows a migration script. Most are shell wrappers around sqlite3 with no record of what ran.
Approach. A single binary with embedded SQL, a checksum per migration, and one transaction per file. It refuses to run if an applied migration changed.
Stack. Go, SQLite, embed.
Result. One binary, one SQLite file, no migration service. The same command works locally and on the server.
GoSQLiteCLI
ledger up
-> 0001_init.sql applied
-> 0002_add_users.sql applied
ledger status
0001_init.sql ok
0002_add_users.sql ok
Home lab · 2026
Pulse
Problem. A home lab had metrics in three places and no simple way to see the last hour.
Approach. A single Go binary writes samples to SQLite and renders SVG on the server. No JavaScript, no external chart library, no database server.
Stack. Go, SQLite, SVG.
Result. A dashboard that survives a laptop reboot and needs no maintenance between visits.
GoSQLiteSVG
SELECT bucket, count(*) AS requests
FROM events
WHERE ts >= datetime('now', '-1 hour')
GROUP BY bucket
ORDER BY bucket;
Personal tool · 2025
Archive
Problem. 40k photos scattered across folders; no search by date, camera, or place.
Approach. An indexer reads EXIF, stores it in SQLite FTS5, and emits static HTML. The archive is readable without the tool.
Stack. Python, SQLite FTS5, EXIF.
Result. Search by camera or date in milliseconds; the index is a single file.
PythonSQLite FTS5EXIF
for path in archive.rglob("*.jpg"):
meta = exif.read(path)
index.add(path, meta.taken_at, meta.camera)
# query: camera="X100V" after=2024-01-01
# -> 1,284 photos in 4 ms
Source is available on request. There is no public demo; these tools run on private machines.
Ask for a walkthrough →