DooDooLamb News
Show HN: EchoChain to see what happens when LLMs play telephone
Brief published November 3, 2026 ยท Original source published November 2, 2026
Original reporting by smusali at github.com.
Automated brief. Verify important details at the original source.
What happened
A developer released echoChain, an open-source tool that chains LLM responses in a loop to measure information retention across multiple model exchanges. Each model converts the previous response into a new question, which the next model answers, mimicking the telephone game. An optional baseline install adds sentence-transformer similarity scoring to quantify how much the content drifts. The tool is available via pip and requires an OpenRouter API key to run against hosted models.
Why it matters
Builders using multi-agent pipelines or chained LLM calls need concrete ways to measure semantic drift across hops. echoChain provides a lightweight, reproducible method to surface that degradation, which is otherwise hard to observe without tooling. The sentence-transformer scoring gives a quantitative signal rather than relying on manual inspection.
What to watch
Whether the project adds support for additional embedding models or direct API providers beyond OpenRouter, and whether community benchmarks emerge showing which model combinations retain information most reliably across chain depth.