omdsh-usage

After a long stretch of DeepSeek Harness conversations, the most natural question is how much it all actually cost. omdsh-usage answers in plain sight: session spend, project total, and account balance pinned to the header of the web GUI you are already reading-no log digging, no manual token math.

It ships with the published peak and off-peak rate table (off-peak is half price, covering roughly 17 of 24 hours each day) and records, per model and per call, which tier each token was billed under. Multiple models across multiple tiers are priced at their respective rates rather than blended into one average. All arithmetic runs on the host; the browser receives formatted numbers and never a rate to multiply, so a stale tab cannot render outdated prices. When the balance dips below a configurable floor (default 10 in the account currency), all three icons turn red-a missing credential or a network blip leaves them in their ordinary colour and explains why in the tooltip.

Built for solo developers tracking nightly batch runs, teams that need per-project AI spend attribution, and deployments that want to override the rate table to match a private gateway. It works out of the box with the API key the harness already holds for model requests, and tweaking currency, rates, or the low-balance threshold in Settings takes effect on the next poll-seconds later, with nothing to restart.