The AI gateway for agents. One API key and one balance for language-model inference, image and video generation, and private GPUs at live market prices.
scalarware.com · Docs · llms.txt
Serverless inference across frontier and open-weight models, image and video generation, and on-demand GPU capacity — all behind a single OpenAI-compatible endpoint, one API key and one balance.
0% markup on language-model inference. Publicly, permanently. You pay what the model costs. Revenue comes from GPU rental and media generation, not from a spread on tokens.
Point Claude Code, Codex, Cursor or any MCP-capable client at the docs and it can do the whole loop on its own — pick a model, rent a machine, deploy weights, poll until the model answers, tear it down:
Read https://docs.scalarware.com/llms.txt and follow the instructions.
Deployments report their own setup phase (preparing → downloading → starting → ready), so an agent knows whether the box is booting or the weights are still
pulling, and can tell its user which.
Open-weight models packaged to run on rented capacity with no setup — the server, the weights and the API surface are configured before you connect:
| Template | Task |
|---|---|
| Qwen3.8 27B | Text generation (vLLM) |
| LTX 2.5 | Video generation |
| FLUX.2 Dev | Image generation |
| Krea 2 Turbo | Image generation |
Billed by the minute against your balance. A machine that runs out of funds stops automatically.