A secure, visual interface bridging the gap between standard workstations and the WSU Kamiak High-Performance Computing cluster.
> Establishing SSH tunnel to core.kamiak.wsu.edu...
> Authenticating via identity file...
> Tunnel established on port 8080.
> Fetching cluster status...
> Nodes available: 12 | Queue: Idle
> Routing GUI traffic via proxy...
> Ready.
Eliminate the learning curve of bash scripting and command-line interfaces. Kamiak GUI provides an intuitive dashboard for full cluster management.
Securely connect to the WSU Kamiak cluster via an automated SSH tunneling backend. No manual proxy configurations required.
Design SBATCH scripts visually. Select partitions, nodes, GPUs, and time limits through dropdowns rather than editing text files.
A built-in file explorer to upload datasets, modify scripts, and retrieve results directly from your scratch directory.
Launch and interact with local LLMs (like Llama 3 or Mistral) running on compute nodes via a dedicated chat interface tab.
Kamiak GUI is built as a standalone desktop application using modern web technologies. The frontend is developed in React (Next.js), ensuring a highly responsive and familiar UI.
To communicate with the HPC cluster securely, a specialized backend acts as a bridge. It
leverages SSH2 protocols to establish
encrypted tunnels, executing commands (like squeue or sbatch) on
behalf of the user.
When an LLM is launched, the backend creates reverse port forwards, mapping the model's API running on a remote compute node directly to the local interface.
Next.js UI runs locally on the researcher's machine.
Node.js backend establishes SSH connection to Kamiak login nodes.
Commands are passed to the Slurm scheduler; port forwards map model APIs back to the UI.
Kamiak GUI doesn't just run LLMs; it orchestrates them. Access a suite of built-in agentic tools designed for complex analysis.
Create specialized AI "Gems" pre-prompted with specific instructions and system roles. Tailor an agent for forensic accounting, while another acts as a legal code reviewer.
Upload massive datasets of unstructured text (PDFs, witness statements, case files). The GUI provides a Notebook LM-like interface to directly query and synthesize information across all your connected documents.
When enabled, securely grant your local LLM the ability to search the public web. It seamlessly combines real-time information gathering with local, secure data processing.