You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Reads GET /v1/quota?model=<name> on the gateway of the platform provider (the
origin of its baseURL) for each of its models: at most once per minute and per
model for the whole process, plus once after a model response ends, with the same
floor. Errors back off up to 15 min; a 404 parks the model for an hour.
Display as a state rather than raw percentages: normal or low priority, with the
time back to normal, for a soft budget; daily, hourly and weekly hard caps;
per-minute caps above 80%; requests in flight from 12 of 16.
OpenCode v2 (2.0.25 and later): token from the host connection
(ctx.integration.connection); snapshot and toasts sent over RPC to a TUI part
that adds a status line after prompt.footer.status, toasts on state changes and /quota; a quota tool.
OpenCode v1 (1.17 and later): toasts on state changes (client.tui.showToast) and
the quota tool; token from opencode-keycloak-auth's token source
(patches/opencode-keycloak-auth-token-source.diff).
OPENCODE_QUOTA_LOG_FILE, since OpenCode v2 shows plugin console output nowhere.
End-to-end tests against OpenCode 1.17.11, 1.19.6, 2.0.25 and 2.3.2
(npm run test:e2e).