Thanks to visit codestin.com
Credit goes to github.com

Skip to content

Tags: wjf1/commandcode-proxy

Tags

v4.11.0

Toggle v4.11.0's commit message
v4.11.0 — dashboard hotfix, protocol compat (count_tokens / include_u…

…sage), .env loading, budget alerts, concurrency cap

v4.10.0

Toggle v4.10.0's commit message
CommandCode Proxy v4.10.0 — admin auth, reliability hardening, SPA ex…

…traction, CI

v4.9.2

Toggle v4.9.2's commit message
v4.9.2

Fix: make silent toast failures self-diagnosing. The machine's master
notification switch (ToastEnabled=0) was rejecting every toast with
GlobalSettingDisabled while Show() still succeeded - including ZCode's own
notifications. The switch is now read at startup and in notify(), with the fix
path logged when it is off; registry reads go through PowerShell because
reg.exe command lines are rejected on this machine.

v4.9.1

Toggle v4.9.1's commit message
v4.9.1

Fix: desktop notifications were silently dropped by Windows because the toast
was sent under an unregistered AppUserModelID. Register CommandCode.Proxy under
HKCU so toasts display as "CommandCode Proxy", with a PowerShell AUMID
fallback and 8 new tests.

v4.9.0

Toggle v4.9.0's commit message
v4.9.0

Add: end-to-end throughput/latency distribution per model, a quota burn-rate
projection computed from differencing the official window usage (not local
history, which only sees proxied traffic), and Windows toast notifications for
quota exhaustion, account switching and engine pause.

v4.8.0

Toggle v4.8.0's commit message
v4.8.0

Add: session attribution from the client-declared x-session-id (cross-checked
against the on-disk session directory) and project attribution inferred from
system prompt text, labelled per row by confidence and left empty rather than
guessed. Also fixes header case handling, path truncation on \node_modules, and
a frequency heuristic that could never fire. Day grouping now uses the client
timezone.

v4.7.0

Toggle v4.7.0's commit message
v4.7.0

Add: cache savings visualisation (amount saved by cache hits vs full input
rate, priced at each record's own timestamp) and a peak/off-peak billing
indicator (current rate tier, time until switch, active rates per model, with
the weekend jump expected correctly).

v4.6.0

Toggle v4.6.0's commit message
v4.6.0

Fix: dashboard cost was ~15x the official bill. Cache-hit input was priced at
the full input rate although cache reads cost 1/50 of input and agent traffic
hits cache on 96-99% of input tokens. Peak-hours requests were also understated
by half because the official timeOfDay rates were dropped when parsing.
Cost now adopts the authoritative upstream gateway.cost; local estimation
prices cache read/write separately and selects the rate by time of day.

v4.5.0

Toggle v4.5.0's commit message
v4.5.0 — plan/billing-cycle API, per-plan model availability, real ve…

…rsion reporting