free-gpu on PyPI | GPU compute planner

Get more from free GPU providers. Tune locally, offload the heavy work smartly.

free-gpu sits on top of llmfit to show what your hardware can handle and when it makes sense to move fine-tuning, inference, or longer runs to free and near-free providers.

View repository Read the README Open dataset

pip install free-gpu

61 providers

Example request

{
  "tool": "plan_provider_workflow",
  "arguments": {
    "workload": "finetune-lora",
    "model": "llama-3.1-8b",
    "budget": "under-25",
    "task_hours": 6,
    "min_vram_gb": 16
  }
}

Feasible Implementation

Codex

Add free-gpu to Codex

codex mcp add freeGpu --url https://free-gpu.vercel.app/mcp

codex mcp add free-gpu-local -- free-gpu-mcp

Claude Code

Add free-gpu to Claude Code

claude mcp add --transport http free-gpu https://free-gpu.vercel.app/mcp

claude mcp add --transport stdio free-gpu -- free-gpu-mcp

Cursor

Add free-gpu to Cursor

{
  "mcpServers": {
    "free-gpu": {
      "url": "https://free-gpu.vercel.app/mcp"
    }
  }
}

VS Code

Add free-gpu to VS Code

{
  "servers": {
    "freeGpu": {
      "type": "http",
      "url": "https://free-gpu.vercel.app/mcp"
    }
  }
}

Run free-gpu

TUI - CLI

Start from the terminal

pip install free-gpu
free-gpu ui

HTTP endpoint

Hosted MCP

https://free-gpu.vercel.app/mcp

Local server

Run the MCP locally

free-gpu-mcp

Canonical workloads

scratch-train
finetune-lora
inference
batch-eval
agent-loop

Provider lanes

Burst

Short, low-friction work

Best for quick inference, demos, notebooks, and small agent loops that should not require a long-lived allocation.

Session / heavy

Medium runs and repeated evals

This is where credits, starter plans, and API-friendly providers start outperforming pure free-tier options.

Grant-scale

When the compute stops being casual

For larger VRAM and longer jobs, the planner shifts toward programs, allocations, and grant-style infrastructure.