{"name":"agent-smith","repo":"Chengjun023/agent-smith","url":"https://midorreal.com/project/chengjun023-agent-smith","repository":"https://github.com/Chengjun023/agent-smith","description":"Token 超支？快给史密斯打电话！ Better Call Smith. Adaptive model routing, a glassy Codex usage monitor, and reproducible benchmarks with receipts.","facts":{"license":{"value":"MIT","source":"https://github.com/Chengjun023/agent-smith","checked":"22 Nov 2026"},"language":{"value":"Python","source":"https://github.com/Chengjun023/agent-smith","checked":"22 Nov 2026"},"releases":{"value":"0","source":"https://github.com/Chengjun023/agent-smith/releases","checked":"21 Nov 2026"},"contributors":{"value":"1","source":"https://github.com/Chengjun023/agent-smith/graphs/contributors","checked":"21 Nov 2026"},"archived":{"value":"no","source":"https://github.com/Chengjun023/agent-smith","checked":"22 Nov 2026"}},"stars":9.0,"writeup":{"what_it_is":"Agent Smith is a tool for optimizing LLM API costs through adaptive model routing and usage monitoring. It includes a local router that selects appropriate models based on task difficulty, and a native macOS floating window that displays token usage and cost comparisons.","audience":"Developers using Codex who want to reduce LLM API spending.","claims":[{"kind":"specific","claim":"Router achieved 86.89% lower reference cost than fixed Astra model in pilot benchmark","excerpt":"Against fixed Astra, the router used **86.89% less reference cost** and **0.75% fewer actual tokens** in this pilot.","status":"not_checked"},{"kind":"specific","claim":"All six pilot tasks passed with local router, fixed Astra, and fixed Sol models","excerpt":"Six public synthetic tasks × three strategies × one attempt: **18/18 passed** the deterministic graders.","status":"not_checked"},{"kind":"broad","claim":"Provides reproducible benchmarks with detailed cost receipts","excerpt":"Every claim comes with receipts.","status":"too_broad"},{"kind":"broad","claim":"Routes routine work to lighter models and difficult cases to more capable models","excerpt":"Routine work gets a lighter model; difficult cases bring in the heavy hitters.","status":"too_broad"},{"kind":"specific","claim":"Float window requires macOS 14+ and displays task usage with token counts and cost savings","excerpt":"Float requires macOS 14+, `/usr/bin/python3`, and Xcode Command Line Tools.","status":"not_checked"}],"alternatives":[],"written_by":"AI, from the project's README","date":"2026-10-11"},"why_now":null,"owner_supplied":null,"score":{"verdict":"mid","hype":2,"reality":9,"gap":-6,"momentum":null,"confidence":60,"version":"score-2.0","rule":"verdict-2.0","calculated_at":"2026-11-22T08:00:00+00:00","verdict_source":"rule"},"early_read":null,"changes":[{"date":"2026-09-30T15:58:32+00:00","text":"Repository created","source":"https://github.com/Chengjun023/agent-smith"}],"cite":"agent-smith is a GitHub project at https://github.com/Chengjun023/agent-smith written mainly in Python, described there as \"Token 超支？快给史密斯打电话！ Better Call Smith. Adaptive model routing, a glassy Codex usage monitor, and reproducible benchmarks with receipts\".","attribution":"Data from Mid or Real, https://midorreal.com/project/chengjun023-agent-smith, checked 22 November 2026. Attribution with a link is required.","still_real":null}