AI agent chaos ends here.
I’ve chased Silicon Valley hype for two decades, watching devs drown in tools promising the moon. And now? AI agents everywhere—Claude Code, Cursor, Codex—each demanding its own Frankenstein setup of skills and MCP servers. Manual installs, copy-pasted docs, zero versioning. New teammate? Hand-hold city. New machine? Repeat hell. Sound familiar? Yeah, the original post nails it.
The more AI coding tools we adopted, the messier our setup got. Skills and MCP servers installed manually, via individual commands, or copy-pasted from docs. Scattered across Claude Code, Cursor, Codex, Windsurf - with no way to version any of it, no way to share it with teammates, no way to reproduce it on a new machine or project.
That’s the pain. Kasetto—pronounced like cassette, that old-school tape that just worked anywhere—steps in with a Rust-forged hammer. One YAML file declares your entire empire: agents, skills from GitHub (or GitLab, Bitbucket, even self-hosted), MCP servers. Commit it to repo. kst sync. Boom. Identical setup everywhere. No drift. No excuses.
Here’s the cynical vet take: We’ve been here before. Remember Cargo revolutionizing Rust deps? Or Nix forcing reproducible builds when everyone else was npm hell? Kasetto’s that for AI agents—what uv did for Python, but cross-agent, declarative, zero-runtime-deps. Single binary. Fast. CI-ready with --dry-run and --json. It’s not hype; it’s the quiet fix we’ve needed since agents went mainstream.
Why Your AI Setup Sucks (And Kasetto Doesn’t)
Look, you’re probably juggling five agents right now. Claude for code reviews, Cursor for editing, Copilot lurking in VS Code, Gemini CLI for quick hacks. Each has its “skills”—fancy word for plugins—and MCP servers for… whatever MCP does (multi-cloud something? Who cares, it merges ‘em automatically). Install one? Hunt GitHub, clone, tweak paths, pray.
Team grows? Onboarding’s a ritual sacrifice. “Run this curl, edit that JSON, trust me bro.” Change a skill? Finger-crossing propagation. Kasetto laughs at that. Global or per-project scopes. Lockfiles. Interactive kst list. kst doctor for diagnostics. Even kst clean to nuke it all. Short alias kst—because nobody types novels.
And the presets? 21 built-ins. Claude Code to Windsurf. Pulls from anywhere, installs to agent dirs, fuses MCP into native settings. Next sync? Delta updates only. Efficient. Rust ensures it flies—cold starts that don’t make you brew coffee.
But wait—cynic mode: Who’s monetizing this? Nobody. Open-source, solo dev (pivoshenko). Stars fuel it, not VCs. Roadmap’s agent management, hooks. No SaaS pivot in sight. Rare in AI land, where everything’s a “platform” begging for your data.
Does Kasetto Replace Your Dotfiles Ritual?
Not quite—but it should. Dotfiles gave us declarative machines; kasetto does agents. YAML like this:
agent:
- claude-code
- cursor
skills:
- source: https://github.com/org/skill-pack
skills: "*"
# etc.
kst sync. Done. Portable as a cassette tape—record once, hand off, identical playback. Historical parallel? Think early Docker Compose for services. Before Kubernetes bloat, it was YAML magic making containers team-ready. Kasetto’s that purity for AI. Prediction: In six months, every AI-heavy repo commits kasetto.yaml. Or teams stay in manual purgatory.
Install? Curl script, Homebrew tap, Cargo. macOS/Linux. Completions for bash/zsh/fish. kst self update. Unhinged simplicity.
Skeptical? --dry-run everything. JSON for scripts. Non-zero exits. It’s devops-minded, not toy.
One quibble: 21 presets solid, but enterprise GitLab? Self-hosted? Covered. Still, if your agent’s obscure, hack the YAML. No hand-holding UI—command-line purist.
The Money Question: Who Wins Here?
Devs. Teams. That’s it. No cloud lock-in. No “freemium” gotchas. AI companies? They hate reproducibility—it means less fiddling, fewer support tickets. But users? Time saved is code shipped. Onboarding drops from days to minutes. Imagine: Junior dev clones repo, kst sync, productive hour one.
Unique insight—the PR spin dodge: This isn’t “AI revolution.” It’s infrastructure maturing. Like when Git fixed version control chaos in 2005. Agents are the new code; kasetto’s the plumbing. Ignore it, watch setups rot as adoption spikes.
Roadmap teases agents/hooks. Issues welcome. Drop a star if it clicks.
**
🧬 Related Insights
- Read more: Pardonned.com Scrapes US Presidential Pardons Into Something Usable
- Read more: Cloudflare Cracks the Code: ASTs Turn Workflow Scripts into Stunning Visual Maps
Frequently Asked Questions**
What is kasetto used for?
Manages AI agent environments declaratively—skills, MCP servers across 21+ agents like Claude Code and Cursor. YAML config, sync once, reproducible everywhere.
How do I install kasetto on Mac?
brew install pivoshenko/tap/kasetto or curl the install script. Cargo works too. Single binary, no deps.
Does kasetto work with team repos?
Yes—commit YAML to GitHub/GitLab. kst sync pulls/installs identically. Global/project scopes, lockfiles prevent drift.
