Butterfly traces multi-agent failures with graphs, 62.5% accuracy vs 53.8% for log prompting
JeremyCMorgan · x · 2026-11-04
Developer Pulkit Agarwal built Butterfly, an open-source tool from the FalkorDB hackathon that answers "which agent broke it, and at which step?" for multi-agent systems — where failures surface at the end but the first wrong move may be a hundred steps earlier.
How it works:
- Treats a failed run as a dependency graph, not a message list: every message becomes a (:Step) node recording the agent and the facts it introduced (URLs, filenames, numbers, names), stored in FalkorDB.
- Instead of scrolling logs, it traces errors backward through the graph to find the earliest flap of the butterfly's wings.
- A real Who&When benchmark case: a browsing agent clicks an irrelevant site at step 12, the team circles for 15 more steps, and the run dies at step 28.
Results: identifies the responsible agent 62.5% of the time vs 53.8% for whole-log prompting, using half the tokens — with honestly stated limitations.
- Demo: butterfly-falkordb.netlify.app · Code: github.com/PulkitAgrwal/butterfly
More from coding & agent
- Veteran game dev: "AI builds a full game in a day" posts are mostly hype — bendee983 · 2026-11-04
- Dev builds interactive Excel-to-dashboard with ChatGPT, shares the prompt — TermRemarkable665 · 2026-11-04
- Agents cut dev time from weeks to a day; bottleneck moved to specs and review — Suspicious_Orchid770 · 2026-11-04
- Orgo, a YC-backed platform giving AI agents their own cloud desktops, claims 1,000+ teams — Scobleizer · 2026-11-04
- Founder: Vibe-coded a PMF-successful product, then hired pro devs to fix the mess — sujingshen · 2026-11-04
- One prompt to audit cross-region traffic: dev cut trans-Atlantic latency by moving a service — DanielLockyer · 2026-11-04