---
title: README
emoji: 🟢
colorFrom: green
colorTo: gray
sdk: static
pinned: false
---

**Scalably** builds and runs AI agents for real businesses. One founder directs AI agents that do the engineering; the owned inference tier runs on two GeForce RTX 5090s.

We publish what we measure: serving benchmarks with every row, research notes with their evidence ledgers, and corrections in the open.

- Datasets: [MoE vs dense on vLLM](https://huggingface.co/datasets/Scalably/moe-vs-dense-vllm-rtx5090) · [Qwen Flash 256K engines](https://huggingface.co/datasets/Scalably/qwen-flash-dual-5090-serving) · [Qwen agent fine-tuning results](https://huggingface.co/datasets/Scalably/qwen-agent-sft-experiments) · [Neural Graph Substrate ledger](https://huggingface.co/datasets/Scalably/neural-graph-substrate-evidence)
- Explorer: [RTX 5090 serving explorer](https://huggingface.co/spaces/Scalably/rtx5090-serving-explorer)
- Code: [github.com/scalably-io](https://github.com/scalably-io) · Research: [scalably.io/research](https://scalably.io/research) · Blog: [scalably.io/blog](https://scalably.io/blog)