Compute & GPU Resources
FleetQ can manage compute resources for AI workloads. Connect GPU providers like RunPod for tasks requiring dedicated compute — fine-tuning, large model inference, or batch processing.
Compute providers
| Provider | Description |
|---|---|
| RunPod | On-demand GPU instances (A100, H100, etc.) with pod lifecycle control. |
| Vast.ai | Marketplace GPU rentals — typically the cheapest option for batch jobs. |
| Fal | Serverless inference for image, audio, and video models. |
| Replicate | Hosted model inference API (Stable Diffusion, LLaMA variants, custom models). |
| Browser sidecar (Xvfb) | Headful Chromium + patchright running inside an Xvfb virtual display. Use for anti-bot scenarios (Reddit, Cloudflare) that block headless browsers. Combine with a proxy credential for residential-IP bypass. |
| Bridge | Use your own hardware via FleetQ Bridge (local LLMs, local agents). |
| Platform | Default shared compute for standard LLM API calls. |
CostEstimator projects spend per compute job across all
providers before scheduling — useful for budget gating and plan-limit enforcement.
RunPod integration
Manage RunPod pods directly from FleetQ. Create, start, stop, and monitor GPU instances. Configure templates, volume mounts, and environment variables for each pod.
# Manage RunPod pods via MCP
runpod_manage(action: "list")
runpod_manage(action: "create", template_id: "...", gpu_type: "NVIDIA A100 80GB")Bridge compute
FleetQ Bridge connects your local hardware to the platform. Run Ollama, LM Studio, or any local LLM endpoint and use it as a compute backend for agents and experiments — at zero API cost.
MCP tools
| Tool | Purpose |
|---|---|
| compute_manage | List and inspect connected compute backends |
| runpod_manage | Create, start, stop, and monitor RunPod GPU instances |
Cost tracking
GPU compute costs are tracked in the budget system alongside LLM API costs. RunPod spend is recorded in the credit ledger and counts toward team budget limits and alerts.