lidge-jun/opencodex v2.6.0 on GitHub

What's New in 2.6.0

A full usage tracking pipeline, from raw request logs to aggregate API to visual dashboard:

Persistent usage log — every proxied request is logged to ~/.opencodex/usage.jsonl with token counts, latency, and provider metadata
/api/usage endpoint — aggregate summary with day-grid, per-model, and per-provider breakdowns
Usage tab — GitHub-style activity heatmap expanded to a full-year contribution grid
Dashboard token stats — 30-day token totals and coverage ratio on the main dashboard
Logs token column — inline token counts in the request log table
ChatCompletions usage shape — parse prompt_tokens/completion_tokens alongside the Responses shape
SSE Content-Type detection — correctly detect SSE streams when upstream omits the Content-Type header
Provider pool merging — merge codex pool log-label suffixes in summary aggregation
Usage debug capture — OPENCODEX_USAGE_DEBUG=1 dumps upstream response shapes for troubleshooting

ocx status --json — machine-readable runtime diagnostics (port, pid, provider state)
Help surface split — ocx help and ocx status extracted into dedicated modules
Runtime port reporting — status shows the actual fallback port when the default is taken

Go/Free 30d-only quotas — normalize quota display to monthly-only for Go and Free plans
configuredPlan forwarding — pool quota fetch receives the plan context for accurate bucketing
Service tier in logs — show the fast/standard tier used per request

Service lifecycle fixes: stop-before-cleanup ordering, scheduler XML task install, stale PID cleanup
Passthrough: native SSE body handling, response close reason recording, stream rewrap avoidance
Runtime: bun diagnostics exposure, websocket idle policy, service wrapper startup logging

Published to npm as @bitkyc08/opencodex@2.6.0 (latest).
npm install -g @bitkyc08/opencodex