What's new
- opencode go is LIVE — read straight from opencode's own usage endpoint (
GET opencode.ai/zen/go/v1/usage, authenticated with theopencode-gokey already in~/.local/share/opencode/auth.json). Same numbers as the opencode.ai console:- Account-wide — every machine's spend folded in server-side, no SSH fan-out needed
- Exact resets — each window states its own reset moment, so the countdowns are the server's rather than an inference
- The server's verdict — a window it calls
rate-limited/limit/blockedreads as walled whatever the percentage says, and is marked the binding constraint - Polled at most every 2 min, cleared by a manual refresh; a last-good reading serves for 15 min if the endpoint goes quiet
- The opencode.db estimate is now the fallback, for a machine with no Go key or no route to opencode.ai — anchored 5-hour block, billing-cycle month, Go's model multipliers, and other machines over SSH, all unchanged and labeled EST
- Monthly ceiling is configurable —
"go_monthly_cap"in~/.config/tokenmaxxing/config.json(defaults to the $40 promotional cap; set60for the published one) - Compact rings carry the reset next to the dollars now that opencode's resets are trustworthy
- Grok refreshes under the Grok CLI's own auth lock, so the two clients stop spending each other's single-use refresh token and revoking the session; an
invalid_grantrefusal now reads as "run `grok login`"
Builds
| Platform | Stack |
|---|---|
| KDE | Rust · GTK4 · libadwaita · Cairo |
| macOS | SwiftUI · Liquid Glass |
Build from source — see the root README and each platform's README.