cohix

Sign in Subscribe

Latest — 05 Nov 2026

Maximum-security agent containment with microVMs: awman v0.13

A core rule of awman has always been that agents never run on your host. This meant installing something else before you could use awman: Docker, Apple Containers, or Docker Sandboxes. That's one more daemon to keep running and update, and second set of tools you need to

More issues

I created a squad of agents to do all the boring work for me (awman v0.12 is out)

Much of my day building software is not the fun, interesting stuff like writing a new feature or working on an interesting performance optimization. It's the issue triage, dependency and pipeline management, PR merge conflicts, and the work item that needs a rough plan before anyone else can

Dynamic workflows with any harness, any model: awman v0.11

I’ve been away from working on awman for a few weeks now. I finished paternity leave and went back to work, so needed a while to settle in! I don’t expect weekly releases every single week going forwards, but the project IS still under very active development, just

A penny for your context?

Welcome back to the PrettySmart show after a week on Vancouver Island exploring what may be the most laid back part of Canada I've witnessed yet: Tofino, British Columbia. This week on the show we have several very different but long-awaited new pieces in awman: Experimental support

Workflows from issue to merged PR: awman v0.9 is out

In the last post, I announced that amux was becoming awman — the Agentic Workflow Manager. The rename reflected a shift in focus: the tool has grown from a terminal multiplexer for agents into something closer to a software development lifecycle pipeline tool, orchestrating the entire process from spec to merged

amux is becoming awman: an Agentic Workflow Manager for the entire software development lifecycle

When I started building amux earlier this year, the name stood for "agent multiplexer." The idea was simply "tmux, but for agents running in containers", which was immediately useful for me in my daily work. As my "agentic engineering" proficiency grew, the tool grew

Behold the miraculous self-improving codebase! ane v0.2 is out.

Since one of the core goals of ane is to allow code agents to explore, read, and edit source code with one-shot commands using chords and a simple CLI, this week I experimented with a self-improving codebase to work on ane from a code agent's point

Introducing ane: A language server-aware chord-based terminal code editor (for humans and agents)

TL;DR ane is a terminal code editor that combines a new chord grammar with language servers to allow both human developers and code agents to efficiently explore and edit source code - ane exec lets agents and scripts make precise reads and edits, while ane’s TUI editor lets developers

That refactor sucked, but amux v0.8 is out now

The past few weeks has been spent rebuilding amux's internals from scratch. The previous architecture was mostly “designed” by Claude and Codex across sixty-plus work items, and by the end it had become clear that the three frontends (CLI, TUI, headless) were each implementing business logic in

Code agents are bad at Software Architecture - for now

Hello from paternity leave, week three. There wasn’t an amux release last week because I've gotten so fed up with the patchwork design of the amux codebase I've decided to burn it all to the ground, more or less. Even top-end foundation models like