/ 01 — The problem
What was broken.
Every team building LLM features kept reinventing the same five components — streaming text, citation popovers, stop-buttons, tool-call cards. Patterns weren't documented; bad implementations spread fast.
←Back to work
/ 03stream-ui·2024 — present
Open source library · 2024
stream-ui.
Open-source React primitives for streaming LLM responses — token-by-token rendering, citations, stop-buttons, and the patterns I learned the hard way.
// previewstream-ui
GitHub stars4.2k
teams using in prod~30
licenseMIT
runtime deps0
/ 02 — The approach
How I tackled it.
Extracted the patterns we'd refined inside Stellar Labs and shipped them as a tiny, dependency-free library. Wrote each primitive to be replaceable — you take the bits you want, leave the rest. Focused docs on the WHY (uncertainty, interruption, latency) not just the HOW.
/ 03 — The outcome
What shipped.
4.2k stars within a year. Used in production by ~30 teams. Pattern guide cited in two conference talks I didn't give.
/ Like what you see?