TokenLedger

If you've used DeepSeek Harness for a while, you've probably hit the same awkward moment: the panel shows one cold total token count, but which relay site actually charged you for that, and which project burned it? When you're juggling multiple relay sites with different rate plans, balances, and subscription windows, reconciliation is pure guesswork. TokenLedger is built to close that gap—it attributes every unit of usage to the relay site that truly served the request, names the station by its domain, merges multiple keys from the same site into one line, and requires zero configuration and no extra credentials.

Its core capabilities all serve one goal: make the numbers honest. It auto-discovers relays from host provider configs, groups usage by the working directory where each session started (so subdirectory sessions still count), reads balances and subscription quotas across DeepSeek official, New API, Sub2API, GLM, OpenRouter and more, and estimates cost using rate tables with effective-dating and tiered buckets. It even handles the easy-to-miss cases: failed requests that still billed, duplicate usage reports, and orphaned usage chunks. Rows that can't be attributed are shown explicitly—never disguised as direct-to-vendor traffic.

If you're a heavy Harness user, a small team lead, or a developer maintaining multiple relays and projects, TokenLedger gives you a clear view of where the money went. It never reads prompts or response content—only counters and identifiers—and its endpoints are loopback-only with anti-spoofing built in. Once installed, you don't need to fill in a single field to see the full breakdown; the only feature that needs an API key (balance checks) reuses the one your host config already has.