Skip to content

Bruce Hart

News

Updates, launches, and short notes.

Short updates when I ship something, learn something, or accidentally do both at once.

News items

Launch Jan 27, 2026

harlite: query HAR files with SQL

I made HAR files queryable.

I keep running into HAR files when debugging front-end stuff and when asking a coding agent, "what actually happened on the wire?" The capture is useful, but it is also a giant JSON blob.

So I built harlite: a tiny CLI that imports HAR (HTTP Archive) into SQLite so you can query web traffic with SQL.

HAR is a great capture format, and a terrible interface

HAR is a snapshot of browser network activity. It is also a pile of nested arrays and optional fields. Most workflows devolve into jq one-liners you regret later.

If you can express the question as SQL, you can reuse it, diff it, and hand it to an agent without teaching it your JSON shape.

SQLite makes the questions obvious

Import once:

harlite import capture.har

Then ask normal questions:

sqlite3 capture.db "SELECT url, status, time_ms FROM entries WHERE status >= 400 ORDER BY time_ms DESC LIMIT 20"

Agents already speak SQL. Humans do too. It is a surprisingly good shared language for "find the weird stuff in this trace".

Response bodies are optional, and deduped

By default, harlite only imports metadata (URLs, headers, timings, status codes). If you opt in to storing bodies, it deduplicates them with content-addressable hashing (BLAKE3), so repeated bundles do not explode your database.

This is a tradeoff: you get speed and small size when you want it, and deeper analysis when you need it.

You can always export back to HAR

I did not want a one-way conversion. harlite can export a database back to HAR, with filters (host, status, method, date range, and more). That makes it easier to slice a huge capture into something you can share or attach to a bug report.

If you try it, tell me what hurts

The repo is here: https://github.com/brucehart/harlite

If you use HARs with agents, I would love examples of the questions you keep asking and the schema you wish you had.

Preview Jan 24, 2026

Glimpse of OpenAI Running Cerebras?

I think I got a glimpse of OpenAI testing their new Cerebras chips and it is incredibly fast. Rendered an entire page of text in ChatGPT in under a second. Can't wait to see this in Codex! Exciting times ahead!

Fun Stuff Jan 17, 2026

New Prototype - Poker Boom

I've always had a concept for a small game in the back of my mind, so I used Codex to build a prototype. It's called Poker Boom. It's kind of like a Candy Crush type game except you build poker hands. Still need some tweaking to make it more fun and addictive. But the bones are up in a Github repo: https://github.com/brucehart/pokerboom . Once I'm a little more happy with how it plays, I'll make a Cloudflare Worker project for it and make it available to play in a browser or on mobile. In the meantime, I'd love to hear feedback from anyone that wants to give it a try. Download the repo, run python -m http.server in the folder and visit localhost:8000 to try it locally in the browser.

Work Jan 5, 2026

Celebrating 10 years at PreTalen/Centauri/KBR

Ten years ago I joined PreTalen when we were seven people in a tiny office with a couple cubicles and a conference room. Since then we grew to 100 employees, sold to Centauri, and then sold to KBR. We are now part of a publicly traded, multi-billion dollar company. It has been a wild ride and there is more change ahead as our KBR division is being spun out into a new company this summer (name coming soon!).

Launch Dec 31, 2025

The blog is live!

I will be sharing practical notes on AI, automation, and building software with real tradeoffs.

Wishing everyone a happy, safe, and productive 2026. Exciting time to be alive!