A local-first toolkit for composing coding agents, command compression, durable memory, and verification loops into one calm engineering cockpit.
1.8k
stars
42k
runs
180ms
latency
Hong Phuc builds developer tools, automation systems, and product-grade workflows for the new AI engineering stack. The work is technical, local-first, and designed to keep quality visible.
Mode
Local-first AI
Focus
Agents, DX, infra
Signal
Ship quality
$ codex run ship-agent-workflow --verify
lint: clean tests: 142 passed build: 31s
$ rtk gain --history --scope personal-os
signal extracted: 18 decisions, 4 risks, 3 follow-ups
Project cards are built around engineering depth: constraints, runtime surfaces, quality loops, and the practical signal behind the artifact.
A local-first toolkit for composing coding agents, command compression, durable memory, and verification loops into one calm engineering cockpit.
1.8k
stars
42k
runs
180ms
latency
A daily technical intelligence feed that watches OSS motion, summarizes project deltas, and turns noisy GitHub activity into useful engineering context.
120
repos
daily
briefs
-76%
noise
Containerized creative automation with separate CLI and GUI surfaces, persistent outputs, and reproducible runtime state for fast experimentation.
2
services
shared
image
bind-mounted
state
A tiny, fast Vite experience cleaned up for real-world maintenance: dependency hygiene, stable Playwright checks, and careful UI behavior.
clean
audit
e2e
checks
lean
bundle
The center of gravity is still craft: simple boundaries, explicit tradeoffs, fast feedback, and tools that make good behavior easier.
01
I like interfaces that make hard work feel understandable: sharp boundaries, useful defaults, observable behavior, and escape hatches when the abstraction leaks.
02
The best AI workflows keep the engineer in the loop. Agents should compress toil, preserve intent, and leave a trail that future-you can trust.
03
Developer experience is not polish at the end. It is architecture, naming, ergonomics, feedback loops, and the discipline to delete what does not earn its place.
Markdown-backed essays ship with static metadata, RSS output, and focused reading pages.
Terminal setup, AI workflows, local LLM experiments, and quality gates are treated as one system.
I like projects with hard constraints, clear taste, and enough technical depth to reward careful engineering.