← Back to Discover

Sipcode

Open-source command-line tool that keeps Claude Code sessions lean by flagging context bloat, catching repeated file reads and reporting spend; it also analyses Codex usage logs.

Command-line toolsObservability & evals
Visit website ↗

What it does

Sipcode is a free, MIT-licensed command-line tool that sits between you and Claude Code. After npm i -g sipcode and sipcode init, it installs a hook and reply rules so the next Claude Code session uses it automatically. It watches the context window and warns about what it calls context rot, catches duplicate file reads and other waste before it reaches Claude, and gives measurable receipts in the terminal. Commands report drift from your normal usage, daily spend, a month-end forecast, per-session rewrite stats, and a check for regressions that arrive with an update. For OpenAI Codex it offers spend analytics and usage-limit reports from local session logs, without changing what Codex does. The project reports a 62.6% median saving on its own locked 20-task benchmark, but says that corpus is synthetic and real results depend on the workload. It makes no network calls in normal use and includes an MCP server for querying your own usage.

Capabilities

  • Hook that dedups repeated file reads and caps verbose tool output
  • Context drift and bloat detection with the sipcode drift command
  • Daily spend, trend and month-end forecast commands
  • Regression check across Claude Code and Codex versions
  • Codex spend analytics and plan-limit tracking from local logs
  • Runs locally with no network calls in normal use

When to use it

  • Long Claude Code sessions where cost climbs and answers degrade
  • Tracking Claude Code and Codex token spend and plan limits
  • Checking whether a Claude Code or Codex update changed cache or limit behavior