TokenSpeed Monitor – Locally measure AI coding assistant speed

Real-time token generation speed metrics for Claude, Codex, and other AI assistants to detect throttling

Problem

Developers complain that AI coding assistants (Codex, Claude) sometimes feel fast and sometimes slow, suspecting throttling, but there's no tool to objectively measure it. One user asked for local software to monitor token speed, indicating a widespread frustration; paying for a subscription without visibility into actual performance feels bad.

Solution

A locally running proxy or plugin that intercepts and logs communication with AI coding assistant APIs, calculating tokens per second, time to first token, and total response time. Displays real-time data in a dashboard or menu bar widget, with historical comparison and exportable reports.

Users

Developers who heavily use AI coding assistants, especially tech leads managing team subscriptions and monitoring performance/cost.

MVP

Capture API requests and responses via a local proxy; parse and compute basic metrics like tokens/s and latency; provide simple real-time charts and logs; support CSV or JSON export; compatible with common assistants (Codex, Claude Code, etc.).

Revenue

Free personal version; team version with centralized monitoring, alerts, and reports for $3-5 per seat/month.

Why now

AI coding assistant subscriptions are expensive, and developers are increasingly sensitive to throttling and performance fluctuations, yet transparency tools are lacking. An indie developer can quickly build a local proxy without any cloud dependency.