all projects
001InfraRustEdge2025

Halcyon Edge

A global edge runtime for low-latency AI inference.

[01] SYS.COREuptime 99.992%
RoleFounder · Principal Engineer
ClientHalcyon Labs
Year2025
StatusLive
/ context

Handles 14B requests/mo across 18 regions with sub-40ms cold starts. Built in Rust on top of WasmEdge; ships an SDK in TS / Go / Python.

What started as a private experiment grew into the work I'm now best known for. The technical premise was simple — most teams over-build the runtime and under-build the interface — and the response from early users confirmed it.

What we shipped

A complete rewrite of the runtime in Rust, a CLI that ships as a single binary, and a control plane that runs on three machines without breaking a sweat. The full system fits inside what a normal team would call a prototype.

"The fastest deploy console I've used. Feels closer to a Vim plugin than a SaaS product — and I mean that as the highest compliment."

Eka Setiawan — Staff Eng, Anthemic

Outcomes

  • 01Reduced p99 cold-start latency by 62%.
  • 02Cut infrastructure cost per request by roughly 4×.
  • 03Onboarded 14 paying teams in the first quarter after launch.
  • 04Open-sourced the runtime under MIT in 2025.
← previous

ASCII Studio

Art toy

next →

Shell.fm

Devtool · TUI