What it does
Oh My Sol (OMS) runs on your machine and uses the model access you already have. You describe the work, the model edits files in a separate worktree for the task, and when it thinks it is finished it can only propose that. A checker that the model cannot touch then runs your requirements against a fresh Surfpool chain, and the task closes only if they pass. While it works, the model has Solana references, program error decoding and toolchain checks close at hand, which matters because models were trained on older Solana and tend to get the newer details wrong. You talk to OMS in a chat view, and you can switch to the Workbench when you want the files and a shell in front of you. Once OMS is open, run/connect to add a model provider and /model to pick one, then type a request. Nothing is sent to a model until you do. The quickstart covers each step and what to do if one fails.
Your keys stay yours
On macOS, every command OMS runs goes through a Seatbelt sandbox. It blocks reads from your home folder, apart from the toolchains OMS needs, and it always blocks keypair files and folders like~/.ssh and ~/.config/solana. Security explains each protection and which ones have tests behind them.
Where to go next
Quickstart
Install, launch and connect a model.
How a task runs
The loop from your request to a closed task.
The independent checker
What “verified” means, and what it does not mean.
Slash commands
All 43 commands, grouped by what they do.
Security
What each protection does and which parts have tests.
Benchmarks
Results, methods and limits, with the tables.