Zirn v1.1 — Mizan

Every version of Zirn gets a name.

Version 1.1 is called Mizan.

Mizan is an Arabic word meaning balance, measure, and stewardship. It felt like the perfect name for a release centered around efficiency. As Zirn continues to evolve into a workspace that understands knowledge rather than simply storing it, I've found myself thinking less about adding more AI and more about making the AI that already exists work smarter.

Mizan is the result of that philosophy: less waste, better memory, lower costs, and a noticeably faster experience.

The centerpiece of Mizan is a new memory-aware generation system. Traditionally, AI workspaces regenerate understanding every time a workspace is opened, even when almost nothing has changed. Zirn now takes a different approach. Before generating a new Home page, it evaluates what it already understands and decides whether fresh reasoning is actually necessary. Behind the scenes, Zirn uses a similarity analysis built on the Dice-Sørensen coefficient to compare the current state of a workspace against its previous understanding. Rather than asking whether two versions are identical, it measures whether they are meaningfully different. If the underlying knowledge remains effectively unchanged, Zirn reuses existing understanding instead of rebuilding it from scratch.

The impact is substantial. In testing, a skipped Home generation saved between 2,000 and 8,000+ tokens per interaction, with larger workspaces regularly exceeding 10,000 tokens saved. For users who frequently revisit the same workspace throughout the day, this can translate into an estimated 30-70% reduction in Home-generation token costs. The result is fewer unnecessary model calls, lower operating expenses, and significantly faster access to information that has already been understood once. Instead of paying for the same reasoning repeatedly, Zirn now spends its resources on genuinely new information and meaningful changes.

Mizan also introduces one of the largest performance overhauls Zirn has received to date. Navigation has been redesigned to reduce friction when moving between notes, sources, and workspace views. Page caching has been expanded to dramatically reduce reload times and avoid unnecessary processing. AI reasoning workflows have been streamlined to improve responsiveness while preserving context quality. Collectively, these changes make Zirn feel substantially faster and more fluid, especially in larger knowledge bases where every second of delay compounds over time.

Another addition focuses on continuity. Zirn now remembers the last workspace you were actively using and automatically returns you there when the application launches. Instead of reopening the app and rebuilding your mental context from scratch, you can continue exactly where you left off. It's a small feature on paper, but one that removes dozens of interruptions over the course of a week. The less time spent restoring context, the more time spent actually thinking.

This release also includes dozens of user experience refinements throughout the application. Menus have been simplified. Interactions have been refined. Navigation paths have been shortened. Loading behavior has been improved. Small visual inconsistencies have been cleaned up. Workflows that previously required multiple steps have been streamlined into fewer actions. Individually, none of these changes are headline-worthy. Together, they make Zirn feel more deliberate, cohesive, and polished.

One of the recurring themes while building Zirn has been that users rarely remember individual optimizations. They remember how a product feels. They remember whether it gets out of their way. They remember whether it helps them stay focused on the work that matters. Much of the work in Mizan was dedicated to improving those invisible moments—the seconds saved during navigation, the instant restoration of context, the avoided regeneration, the workspace that simply opens where you left it.

From a technical perspective, Mizan is one of the largest efficiency-focused releases Zirn has received. The application now performs less redundant work, issues fewer unnecessary AI requests, and spends more time reusing valuable context instead of recreating it. Faster navigation, improved caching, remembered workspaces, optimized reasoning pipelines, and memory-aware generation all contribute to a system that feels substantially lighter despite becoming more capable.

What's interesting about Mizan is that it isn't really a feature release.

It's a release about restraint.

Most software becomes more expensive as it becomes more intelligent. Most AI products solve performance problems by throwing more computation at them. Mizan takes the opposite approach. It asks a simple question:

“What if the smartest thing a system could do was recognize when it already understands enough?”

That idea now sits at the center of Zirn's evolution.

The best infrastructure is invisible.

The best optimization feels like magic.

And the best memory is the one you never notice because it was already there when you needed it.

Zirn v1.1 — Mizan

The release that taught Zirn not to pay twice for the same thought.

Zirn v1.2 is now available as a signed Mac download.