SGLang
RadixAttention serving runtime favored for reasoning/agents
SGLang is a high-performance serving runtime whose RadixAttention prefix-cache reuse makes it a favorite for agentic and structured-output workloads; DeepSeek and xAI publicly run large-scale serving on it. It competes head-to-head with vLLM and often leads on throughput for shared-prefix agent traffic in 2025-2026.
Core tech
RadixAttention prefix caching
Notable users
DeepSeek, xAI
How it fits the stack
SGLang with what it depends on (above) and what it feeds (below). The figure renders as a crawlable diagram and upgrades to an interactive 3D graph as it scrolls into view.
SGLang in the AI stack. SGLang with its immediate upstream dependencies (top) and downstream dependents (bottom) in the AI value chain. Hover a node in 3D, or read the full relationships below.
Graph data (text) — 5 entities, 4 relationships
- SGLang —depends on→ CUDA / software moat
- DeepSeek-V3 / R1 —uses→ SGLang
- KV-cache & inference memory tiering —uses→ SGLang
- xAI (capital raises) —uses→ SGLang
Depends on ↑ · 1