# AI coding cost tracker: see estimated spend across every coding tool > Source: > Published: 2026-10-26 15:44:39+00:00 An AI coding cost tracker turns the token counts your coding tools already leave on disk into an estimated dollar figure, per day, per tool and per model. [UsageAtlas](https://usageatlas.com) does this on your own computer for Codex, Claude Code, Cursor, OpenCode and several more, and it is free and open source. The figures are estimates. They tell you where the money is going and whether this week looks unusual, but your provider's billing page stays the record of what you owe. ## Why one tool's counter is not enough Most developers no longer use a single assistant. A morning in Claude Code, an afternoon in Cursor and a review pass in Codex produce three separate meters with three separate views. Claude Code's `/usage` command shows a session cost, and Anthropic's docs say plainly that the dollar figure is computed locally at list price and is an estimate. Cursor and GitHub Copilot keep their own dashboards. None of them adds the others in. A cross-tool tracker answers the questions those meters cannot: - What did AI coding cost me this week across everything I use? - Which model carried most of that cost? - Is today expensive compared with my normal day? ## What UsageAtlas shows UsageAtlas is a local-first desktop app for Windows, macOS and Linux. It reads supported local histories and provider state, then puts tokens, requests, models and estimated cost on one dashboard. | What you see | Where it comes from | |---|---| | Estimated cost from tokens | Codex, Claude Code, Cursor, OpenCode, Antigravity CLI and Pi / OMP, where the model price is known | | Token and session history without cost | Muse Code, whose cost stays unavailable | | Allowances and balances | GitHub Copilot, Warp, Kimi Code, Kilo Code, Factory / Droid, Amp and Qoder | Two details matter for cost work. First, the token breakdown stays next to the total: fresh input, cache reads, cache creation and output, split by model. When a day spikes, you can see whether it was a long cache-heavy session or a switch to a pricier model. Second, a model with no known price shows as unavailable, never as zero. A tracker that quietly counts unknown usage as free would make your totals look better than they are. ## How to set it up 1. Download UsageAtlas from the site or the [GitHub repository](https://github.com/SRX9/UsageAtlas) . 2. Open it. Tools with readable local history, such as Claude Code, Codex and OpenCode, appear without an account. 3. Connect the tools that report through an account or key, such as Cursor or Copilot, from the sources settings. 4. Use the history view and the per-tool filter to see cost by day and attribute it. No account is needed for local tracking. An optional free account keeps a cloud copy of your history so you can restore it on another computer. Prompts and responses are not read. ## Reading the number honestly Treat the cost line as a planning figure. On a Pro or Max subscription, for example, Anthropic notes that the session cost in `/usage` is not what you are billed, because usage is included in the plan. On API billing, contracted rates or discounts can also differ from list price. UsageAtlas labels estimates as estimates for this reason. Use it to compare days, tools and models; use provider billing records for accounting and purchasing. If you only use Claude Code from a terminal and want a quick text report, a CLI reporter such as ccusage may be all you need. Once you work across several tools or want history and alerts in a desktop app, UsageAtlas is the better fit. ## Questions people ask **Is the tracked cost my bill?** No. It is an estimate from tokens and known model prices. Check the provider's billing page for the amount you owe. **Does it need my prompts?** No. UsageAtlas reads usage data, not the content of your prompts or responses. **What does it cost?** Nothing. UsageAtlas is free and open source under the MIT license.