Tokentab totals coding agent token costs by model, project and day
Tokentab is a new MIT-licensed CLI that reads the session logs Claude Code, Codex and Gemini CLI already write to disk and totals the token cost by model, project and day, and the repository has picked up 1,139 stars in nine days.
Image: GitHub
Why it mattersA team that pays for coding agents by the token can finally see which project, which model and which day the money went to, without waiting on a vendor dashboard that may not exist at all.
The developer crwdla published tokentab, a command-line tool that reads the session logs Claude Code, Codex and Gemini CLI already leave on disk and totals the token cost, and the repository has picked up 1,139 stars in nine days. The project is MIT-licensed and was created on 7 September 2026.
What it does
Tokentab parses the JSONL and SQLite session files each of these three CLIs writes locally, matches every prompt and response against published model prices, and prints a bill broken down by model, project and day. A Cursor extractor is included as a stub. The README says the tool runs entirely on the local machine, needs no API key, and uses no network access. Its only third-party dependency is the Rich terminal library.
How it looks
The CLI accepts filters by date, project, provider and model, and can print to the terminal or emit JSON for a pipeline. A local web dashboard runs on localhost:4747 for the same data. Model names in a session log are matched with a fuzzy lookup, so a version like claude-sonnet-4-5-20250929 still resolves to a price when the log records only the family name. Sessions are also categorised into rough activity groups, such as coding, debugging, refactoring and testing.
Where it fits
None of the three vendors exposes a per-project cost view for their CLI. Anthropic's console reports totals by API key, Codex reports by ChatGPT account, and Gemini CLI has no dashboard at all. A team that switches between Claude Code on one repository and Codex on another cannot answer the question "which project did we spend the model budget on last week" from any of the vendor tools. Tokentab reads what the CLIs have already written and answers it locally.
The interest here is real: 1,139 stars nine days after the first commit, on a repository whose whole pitch is a report against a local file. That is a lot of engineers finding that they wanted the same thing.
The cost of adopting it is one Python install and giving it the log directories the CLIs already write. There is no server to run, no data leaving the machine, and no vendor integration to authorise. That matters if the reason a team never set this up before was a security review of a spend-tracking SaaS.
One caveat worth stating. The prices tokentab uses are the ones published by each provider, so a team on a flat-rate Claude Max or ChatGPT Pro plan will see a bill that is the retail-token equivalent of what they used, not what they actually paid. That is still the number a lead wants when deciding whether to move a repository off a flat plan onto pay-as-you-go, or the other way. The categorisation into coding, debugging and testing groups is heuristic and depends on the tool names in each session, so treat those groups as a rough guide rather than an audit.
Source
crwdla/tokentab on GitHub, MIT license.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.
