What it does
sidegrade is an open-source command-line tool, run with npx, that reads session logs from Claude Code, Codex, Hermes and OpenCode on your own machine. It prints a report of what your usage costs and which other models could cover the same work for less. It scores candidate models on the Artificial Analysis Intelligence Index and re-prices your actual token mix on each one. The report recommends a model at the same intelligence level and a cheaper step-down within a tolerance you can set. Prices come from providers' public pricing pages and are bundled in the package. The README says it runs locally with no account, API key or telemetry, makes no network calls at runtime beyond fetching the package, and reads only token counts rather than prompts or code. A JSON option gives machine-readable output so a coding agent can run it and interpret the result, and providers can contribute their prices through a pull request.
Capabilities
- Reads local usage logs from Claude Code, Codex, Hermes and OpenCode
- Re-prices your own token mix on each candidate model
- Ranks models by Artificial Analysis Intelligence Index score and price
- Suggests a same-intelligence option and a cheaper step-down
- JSON output for use inside a coding-agent session
- Runs locally with no account, telemetry or runtime dependencies
When to use it
- Checking whether a cheaper model would cover the work you do in coding agents
- Understanding where coding-agent spend comes from beyond raw token totals
- Comparing providers' prices for the same model
