How I Built a Privacy-First Codex Usage Dashboard on macOS

Short summary

The author details the design of UsageCove Local, a macOS dashboard for Codex session log metadata that prioritizes privacy through data minimization. The tool streams JSONL logs line-by-line, retaining only an allowlisted set of fields (token counts, timestamps, model names) while discarding conversation text and tool arguments. Security measures include loopback binding, random session tokens, HttpOnly cookies, Host header validation, and no telemetry, with explicit caveats about what aggregate data can and cannot prove.

• •Streams JSONL logs line-by-line and retains only an allowlisted metadata field set, discarding prompts and responses
• •Binds to 127.0.0.1 with random ports, session tokens, and HttpOnly cookies to minimize local attack surface
• •Explicitly acknowledges limitations: aggregate rankings cannot prove causal relationships between sessions and usage spikes

Generated with AI, which can make mistakes.