What it does

Receipt records the token counts and cost of AI coding tools in a local, append-only ledger and turns them into a line-itemized summary for a branch or pull request. Usage arrives either through a logging proxy that sits between a tool and the Anthropic or OpenAI API, or through importers such as one that reads Claude Code session logs. The README says it keeps token counts, model names, costs and timestamps, and never stores prompts, completions or source code. A GitHub Action, or the receipt post command run from your own machine, writes a single comment on the pull request and updates it on each push. The same data can be read in the terminal or in a local dashboard that shows spend over time, by model and by branch. Budgets per pull request or per day add a bar that turns red when the limit is passed. Further commands estimate how much of a Claude subscription's five-hour and weekly windows a task used, suggest ways to cut wasted tokens, and check whether a long session is close to the point where quality drops. The README labels the plan figures as rough estimates, and dollar figures depend on a bundled price list that you can override. At the time of the page, install means cloning the repository and building it; an npm package is listed as coming later.

Capabilities

Itemized token and dollar cost per model on every pull request
Logging proxy for OpenAI- and Anthropic-style APIs, plus an importer for Claude Code logs
Per-pull-request and daily budgets with a warning bar
Local dashboard of spend by model and branch
Advice and session-health commands for cutting wasted tokens

When to use it

Showing reviewers what an AI-assisted pull request cost
Keeping AI coding spend inside a per-PR or daily budget
Seeing how much of a Claude plan's usage windows a task consumed