Your research
Interactive allocations
Reserve an allocation through MCP and connect your desktop when you need an interactive editor.
On this page
Reserve only what you need
Batch training, monitoring and result inspection work through the researcher MCP. For an interactive shell or remote editor, ask your client to call research_allocate with a name, CPU, memory, time and GPU request plus a unique UUID request_id.
Check your profile
Use your current resource limits and choose a time limit for the editing session.
Reserve an allocation
research_allocate returns a durable native job and connection_url. Follow research_job until its state is RUNNING.
Approve your public key
Open the connection URL in your browser. Sign in with the same Google account, choose your desktop platform and download the existing OpenSSH configuration.
Release compute when done
Use research_cancel_job with the allocation’s job ID. Closing an editor, chat or browser does not release the reservation.
The same workspace and limits
Your shell starts in /workspace. Persistent files are shared with your own batch runs and later allocations. CPU, memory, GPU and time limits apply to the desktop session. External network access is unavailable inside the allocation.
pwd
ls -lah
python3 --version
printf "notes from my allocation\n" > session-notes.txt