Transfer files to and from GPU instances without manual SCP commands. Your AI agent handles uploads, downloads, and file verification.Documentation Index
Fetch the complete documentation index at: https://runcrate.ai/docs/llms.txt
Use this file to discover all available pages before exploring further.
”Upload my training script and config to the training instance.”
The agent transfers files directly:list_instances— finds the target instancefile_upload— uploadstrain.pyto/workspace/train.pyfile_upload— uploadsconfig.yamlto/workspace/config.yamlssh_execute— runsls -lh /workspace/to confirm
Uploaded 2 files to training-run:Both files verified on the instance.
File Size Remote Path train.py 3.8 KB /workspace/train.py config.yaml 512 B /workspace/config.yaml
”Download the trained model weights and the training log.”
The agent retrieves files from the instance:ssh_execute— runsls -lh /workspace/output/to see what’s therefile_download— downloads/workspace/output/model.safetensorsfile_download— downloads/workspace/output/training.log
Files on the instance:Downloaded model.safetensors (1.2 GB) and training.log (48 KB).
File upload and download have a 10MB limit per transfer. For larger files, use
ssh_execute with wget or curl to transfer directly between instances or from external URLs.”Upload a requirements.txt and install everything on my dev-box.”
The agent combines file transfer with command execution:file_upload— uploadsrequirements.txtto/root/requirements.txtssh_execute— runspip install -r /root/requirements.txtssh_execute— runspip list | wc -lto confirm packages installed
Uploaded requirements.txt (28 packages) to dev-box. Installation output:28 packages installed. Environment is ready.
Tools used in this workflow
| Tool | Purpose |
|---|---|
list_instances | Find the target instance |
file_upload | Transfer files to an instance (max 10MB) |
file_download | Retrieve files from an instance |
ssh_execute | Verify uploads, list remote files, run install commands |