Sno Station: One person. A whole engineering team of AI agents.

Sno Station is the workstation for your AI coding agents. Claude Code, Codex and Cursor each run in their own harness. Sno Station is where those harnesses sit side by side and work as one team. Not another agent, not another model: it runs the ones you already use.

Start with work you would otherwise babysit. One agent builds, another reviews, and they settle the findings before the result reaches you. When one hits its usage limit, the other picks the task up. What you correct today is a rule in their skills tomorrow.

To get it, tell your agent:

install sno.ai from GitHub

Your agent installs it, connects every agent it finds on your computer, and asks you two short questions. Then give two agents their first task:

sno reach pair "Fix the login redirect"

One builds, the other challenges it, and you can watch both. It runs on your own computer with Claude Code, Codex, Cursor, Hermes Agent and OpenClaw. No API key. Open source. Prefer one terminal command?

Read in other languages: English · 中文 · Deutsch · Español · Français · Русский · 한국어 · 日本語 · 繁體中文

Hand the work off tonight. In the morning you get results, not three questions.

Three things that happened to us before we built this:

• An agent stopped to ask a question it never needed to ask. We came back seven hours later and nothing had moved.
• An agent handed us fifteen small questions. Thirty minutes of our attention later, it turned out it could have answered every one of them itself.
• One agent tried to wake another in the middle of the night and could not. Both sat there until morning.

Your agents are fast. What is slow is the waiting: for you, for a usage limit to reset, for someone to carry a result to the next window. So the whole workstation is built around one rule: the work does not stop because one agent did.

• A usage limit is not a stop sign. Another agent continues from the exact commit.
• The author does not get the last word. A different agent reviews, in a session that never saw the conversation.
• "Done" counts for nothing. Each success check is answered by a real run.
• You come back to one page. What got done, what is stuck, what is waiting for your answer.

It delivers. You ship.

Star it now. The next time an agent stalls on you at 2 a.m., you will want to know where this was.

Three ways to use it follow: start a builder and a reviewer with one command, run the whole squad through one agent, and approve a lesson tonight that works in the next reply. Each comes with something we saw on our own machines.

I went to bed. They changed shifts, then argued over the code.

sno reach pair "Fix the login redirect"

One agent builds. The other challenges what it built. They settle the findings between themselves, so the work reaches you already argued over. Both run in tmux windows you can open and watch. With two different agents signed in, you get one of each. With only one, you still get the pair in two separate sessions, and the command tells you the review stayed in the same family.

When one runs out, the other clocks in. For a long run, tell the working agent to use rotate-agent. At 2% of its quota it writes a handover and wakes a different agent, which continues from the exact commit. And when an agent's context fills up and compacts, it still knows what it was doing and what already failed.

The other one sees what the first one can't. Ask either agent to peer-review the other's work. The reviewer always works in a fresh session that never saw the conversation, and it is a different agent whenever you have two. A review does not always find a problem. It keeps finding the ones the author could not see, because an agent reviewing itself brings the same habits it wrote the code with.

What we saw, the night of 2026-09-18:

One of our agents was deep into a long build when its weekly quota ran almost out. It did not stop and wait. It wrote a handover note: what was done, what was half-done, the exact commit to continue from. Then it woke a different agent.

One thing went wrong, and it is the part worth reading. The new agent started editing before it had confirmed the handover. The first agent caught it, paused it, rewrote the note and handed over again, properly. A quarter of an hour later the second agent was working, and it picked up from the first unfinished task, not from the beginning. I was asleep for all of it.

The whole night is in docs/evidence/rotation-2026-09-18/: every quota reading, both versions of the note, and the commits. Only identifying details and credential lines are removed, and the folder's own README lists each one. The timings, the mistakes and the corrections are as they happened.

We talk to one agent. It runs the rest.

Running five agents used to mean five terminals, and us in the middle, carrying what one said to the next. That job is over.

This is the beginning of the README. Read the rest at the original. Read the full README ↗