How 10xs works
Start from an idea or from your repo. You own the repo, the plan, and the merge button.
- 01
New project: point at your repo.
Give the repo; connect once from your machine and 10xs reads the code instead of asking.
- 02
Read the facts back; correct what it got wrong.
10xs writes back what it understood about the code, who it is for, what done means, and what it will not do. Correct any fact before work starts.
- 03
Get a plan that fits the code you have.
The plan is written against the real repo: what to change, in what order, what not to touch. Accept, say "not quite", or regenerate.
- 04
The plan becomes a board.
Cards carry their dependencies, with each one small enough for one agent to finish in one attempt.
- 05
Connect your agents with one line.
Paste the connect line into a terminal on your machine. Your own agents (Claude Code, Codex, Cursor, any agent) claim cards; you set how many run at once.
- 06
Watch work, not logs.
Each card shows its attempt, what the agent did, and a conversation you can type into while it runs. If an agent needs you, the card says why in plain English.
- 07
Only checked work comes back.
Every finished card returns as a pull request the gate has run: tests, types, build, the house rules. You accept or send it back with words.
- 08
Repeat tonight.
New feedback becomes new cards; the agents run again; you review in the morning.
