The engine is first-party and has no runtime dependencies. There is deliberately no tokenizer: there is no public local Anthropic tokenizer, so a token count produced by one and labelled “Claude” would be a fabrication — and the session file already reports the true counts, which is strictly better than any estimate of them.
To take a JSON Web Token apart, use tokenpeek. To find secrets and personal data inside a dataset, use scrubjay.
Built by Ben Richardson. More tools and sites at lab.benrichardson.dev. Source at github.com/ben-gy/runledger.