Improved parallelism in self-hosted agent

Ideas Feature Requests
, , ,
1

Feature request for product/service

Cloud Agent (GitHub, Slack, Linear,…)

Describe the request

When using a self-hosted cloud agent (without a pool) as per My Machines | Cursor Docs you cannot specify a branch or worktree to use.

An agent is limited to a single directory, and it’s not possible to work on multiple worktrees in parallel.

An ideal flow could be something like:

  1. When starting a new cursor agent session with a device under My Machines allow selecting remote agent, plus allow selecting a worktree
  2. Cursor handles creating the worktree, and starting the agent on the target machine and giving it access to the worktree directory
  3. Another session could then be started in another worktree at the same time targetting the same remote machine
5

Hey, thanks for the detailed feature request. The flow you described, where Cursor automatically creates and manages worktrees on the target machine, isn’t supported right now. But you can already get parallel sessions on the same machine today:

  • Create worktrees locally ahead of time and run a separate worker for each directory, or register multiple directories by repeating the --worker-dir flag
  • Each worktree will show up as a separate target in My Machines, and agents on the same machine can run concurrently

Self-hosted worker guide, if it helps: My Machines | Cursor Docs

The part about choosing a branch or worktree for self-hosted agents is a fair pain point. I passed your suggestion to the team as feedback. I can’t share an ETA, but feature updates will show up in the changelog. Related threads if you want to add your vote: Self-hosted workers developer workflows and Agents Windows - select to run on other workstations

Let me know if the workaround doesn’t work with your setup.