Auditable by design. Every model call, tool run, and edit hits a local event log before it executes. If it crashes mid-task, it picks up exactly where it left off from that log. No lost work and no re-prompting.

We pointed Muse Spark 1.2 at a kernel optimization task and let it run. 1,000+ tool calls over 24 hours on NVIDIA Hopper. It kept finding substantial improvements well beyond the initial exploration phase.

Muse Code runs specialized background agents that stay active your whole session, so they build up context over time instead of starting from scratch on every task. When a job is big enough, it fans out to separate sub-agents working in parallel in isolated worktrees. Your working copy is never touched. In testing we had it build six features for a game simultaneously with no collisions.

Releasing Muse Code in beta today. It's a terminal coding agent that takes on complete software engineering tasks across large repos: planning changes, writing code, validating the results. Powered by Muse Spark 1.2, a coding-focused model update.

Pricing: It's easy and low-cost to get started. Install Muse Code with one line and you can start on our contributor tier.

Muse Spark 1.2 is our next step as we push toward frontier, with larger, more capable models on the way. Install it, use it, tell us what you think.

impressive Mark! I’ll have to give it a try

I wonder if that'd cause issues with non indempotent operations.

A local event log changes the shape of a coding agent. It makes the run inspectable as a system process, not only as a chat. Resume is useful, but I would judge it by whether a reviewer can trace each tool call, permission, file change and rollback after the task is done.

I Have Figured Out How To Power The Cooling Of New Data Center Constructions, I USE NO WATER, I Also Cover 10-15 Percent Of The Server Energy, Wanting To Make The Most Powerful Data Centers On Earth, Please Help Me Help You Become The Most Powerful AI Company On Earth, Very Serious, Please Just Listen To My Ideas Is All ...

It's funny to see something I came up with over a month ago get implemented in a major frontier lab

Hello @finkd @instagram We don't need new features. Fix your AI and bring back real human support. All my accounts have been suspended for over 2 months, Please help 💔

Lately, I’ve come to the conclusion that working for Mr. Zuckerberg on building metacognitive structures might be the most productive path. After all, it is the only company that doesn’t churn out as much narrative as the others. (If<What if)

State persistence is the biggest bottleneck holding back true AI agents right now. Pick up where it left off without burning tokens on re-prompting? Game changer for long-running workflows.

That's how serious AI systems should be built. An assistant that can recover from failures without losing context feels far more reliable than one that starts over every time. Auditability isn't just for compliance—it improves the user experience too.

That’s nice

Such a nice output

Hey Mark, don't mind me, just shamelessly plugging my book preorder under what's sure to be a banger of a post. Did I mention the main characters name is Muse?!

video unrelated

Auditability is going to be one of the dividing lines between AI that demos well and AI that organizations can actually trust with real work. When leaders can see what happened, why it happened and where ownership sits, autonomy becomes much easier to scale.