Scalattice

Scalattice is a distributed GPU inference marketplace that exposes a curated catalog of open-weight LLMs (Qwen, Llama 3.3, DeepSeek R1, Gemma and a handful of others) behind an OpenAI-compatible API. It sits in the same competitive space as OpenRouter, Together AI and Fireworks, but its distinguishing bet is a two-sided network: on one side, developers get cheap per-token inference; on the other, GPU owners (individual rigs, small clusters, homelab operators) run an open-source agent to sell idle capacity and get paid per token served. The platform handles routing, health checks, security tiering and regional policy enforcement so that requests can be constrained to specific residency zones, which matters for teams that cannot ship prompts to arbitrary anonymous hardware. Because the API surface mimics OpenAI, integration is usually a base_url swap in an existing SDK — LangChain, LlamaIndex, LiteLLM, Vercel AI SDK and anything that speaks the OpenAI schema work with no code changes. Typical workflows include powering agent loops on cheaper open models, running background summarisation and classification at scale, prototyping RAG stacks without committing to a frontier-lab bill, and giving coding copilots a cheap Qwen-Coder backend. For hardware providers, the agent is designed to be idle-friendly: it accepts work when the GPU is free and steps aside for local jobs. Prepaid credits and live usage dashboards give developers a hard spend ceiling, and payout tracking gives providers a transparent view of earnings. The trade-off is model breadth (a small curated list rather than everything on Hugging Face) and the operational reality that distributed marketplaces tend to have more variable latency than a single-provider inference API.