What we build
Operating surfaces, assistants, and training.
Nautilus helps teams run AI systems over time. ROH makes their behavior visible. miri and TalkMode bring AI into real workflows. Vibecamp helps people learn to use these tools well.
Building from Seoul · Researching everywhere
The AI Agent OS
that starts first.
An operating system for AI agents (Nautilus), a live view of what they're doing (ROH), and the tools that ship on both — billed by the second. No quotes, no SKUs.
01Paths
Start with what is
already built.
Each path leads to a product you can explore now: one to run AI teams, one to watch them, one to work beside them, and one to speak to them. Start with the one that fits your work.
/nautilus
Nautilus OS
Run an AI team that does not just execute tasks, but keeps memory, roles, and improvement from one cycle to the next.
Open product ↗/roh
ROH
Watch your AI's decisions unfold live, so delays, mistakes, and handoffs are visible before they become problems.
See how it works ↗/miri
miri.dev
An AI collaborator for real software teams that can understand a codebase and help move work forward responsibly.
Open product ↗/miridev
miridev-cli
Take the same assistant into day-to-day delivery work, from local tasks to automated pipelines and team operations.
Install from npm ↗/talkmode
TalkMode
A voice assistant for Mac that feels immediate, private, and natural enough to use in the middle of real work.
Open TalkMode ↗02Company
Baryon Labs builds tools
for companies that want AI
to work like a teammate.
We make products for running AI work, seeing what it is doing, and putting it into daily practice. The homepage gives the overview. Each path leads to the full product story.
Who it is for
Professionals who need leverage, not novelty.
Founders, operators, engineers, designers, and domain experts use Baryon tools when the work matters enough that delegation, visibility, and control have to improve together.
How to explore
Choose the path that matches your role.
The root stays short. Product pages carry the implementation details, screenshots, installs, and technical depth for the teams that need them.
03Now
Meet in person,
build an agent team.
Two Baryon gatherings are currently open on the root: a legal AI session for lawyers, and a regular weekend meetup for people building with agents. The homepage keeps this short. Registration and details live on Luma.
Legal AI · For lawyers · Baryon Labs
June 19, 2026
Lawyer meetup:
build your first legal agent.
A small working session for lawyers who want to see how an AI assistant can support research, drafting, and internal workflow in practice.
Register on Luma
Regular · Weekend meetup
Every Sunday
Weekend meetup for
agent builders.
A recurring weekend gathering for people using AI in real work. Demos, short talks, questions, and practical conversation around what is actually shipping.
See the schedule
04How it works
Clear on the surface.
Serious underneath.
You do not need to learn the machinery to understand the value. What matters at the root is simple: our tools help teams delegate work, see what is happening, and stay in control.
Delegation
Assign work without losing context.
Give one task to one assistant or many at once, while keeping the work in a shared place your team can still understand.
- Built for teams, not one-off prompts
- Designed for coordinated work
- Details live in the product paths, not on the homepage
Visibility
See what the AI is doing before trust breaks.
Instead of waiting for a bad result, you can inspect progress, handoffs, and decisions while the work is still moving.
- Useful for operators, reviewers, and clients
- Designed to reduce avoidable surprises
- Technical depth sits behind product-specific pages
Control
Adopt only as much complexity as you need.
Some people want a ready product. Others want a deeper operating surface. We keep both paths open without forcing everyone into the same workflow.
- Start with a tool, expand into a system later
- Use the root to choose, not to study everything
- Go deeper only when the work demands it
For technical teams
If you want the implementation, we keep it visible.
curl -fsSL https://baryonlabs.github.io/install.sh | bash
At a glance
Different tools for different levels of involvement.
Hands-on products
Team workflows
Voice interfaces
Training
Explore
Start with the tool that fits your role and your immediate problem.
Evaluate
Open the dedicated page when you need a deeper explanation, demo, or install path.
Decide
Stay high-level at the root, or go technical only when your team is ready to implement.
05FAQ
Frequently asked.
What does "Baryon" mean, and why?
Baryons are the heavy particles — protons, neutrons — that compose everyday matter. We named the lab after the unsexy thing that everything else depends on. The same way our tools work: the agents you see are built on small, durable primitives that already shipped.
Are your products free?
miridev-cli is free and open source on npm. miri.dev has a free tier and a paid plan. vibecamp and ai-native vibecamp are paid cohorts with partial scholarships available — see the application form.
Can I use your stack with my company's existing tooling?
Yes. miri reads any git repository and writes patches as PRs. miridev-cli has JSON-RPC and exec modes meant for CI. baryon-core is MIT-licensed and runtime-agnostic — pair it with your own LLM provider.
Are you hiring?
Quietly, always. If you've shipped an agent in production and we'd recognize the codebase, send us a link and a sentence. Email's at the bottom.
Why both an English and Korean cohort?
Korea has world-class engineers and very few peer cohorts for AI-native building. We run vibecamp bilingually because the lessons travel but the network is local.