When installing the LatchLoop app to GitHub, you choose which repositories you will give LatchLoop access to. LatchLoop does not persistently store your codebase or generate or store embeddings of it. You can revoke access at any time.
When creating a task in LatchLoop, the LatchLoop agent will commit any changes to a new branch in your repository. To prevent accidents, LatchLoop is not able to make commits to your main branch.
Currently, no user tasks or code added to LatchLoop is used to train models. We’re in the business of helping you build great products faster. We are not a model provider. Should this ever change in the future, we would provide customers with the ability to completely opt out.
LatchLoop can create git commits and pull requests in the connected project repository. However, it does not have any other kind of network or application access.
Code read and written by LatchLoop normally interfaces directly with GitHub through our GitHub app. If you choose to use a temporary sandbox when running a cloud agent, your code is downloaded to and temporarily accessed in that sandbox only for the duration of the run. You can also run cloud agents without downloading your code to a sandbox.
There is no way to share code or tasks publicly from LatchLoop. Teams using LatchLoop can share project access to work on tasks with team members, but there is no way for tasks to be viewed by anyone who has not been granted explicit access to your team and project.