
A small Raycast front end for CLIProxyAPI. It covers a subset of the official Management Center: auth files, quota and rate-limit resets.
remote-management.secret-key. CLIProxyAPI hashes it on startup, so keep a copy of the plaintext.http://localhost:8317Every auth file grouped by provider, with:
Switch views from the search bar dropdown or the Switch View action, or cycle through them with ⌘D:
| Action | Shortcut |
|---|---|
| Refresh Account | ↵ |
| Refresh All Accounts | ⌘R |
| Reload Quota | ⌘⇧R |
| Next View | ⌘D |
| Open Management Center | ⌘O |
| Use Rate-Limit Reset (asks first) | ⌘⇧U |
| Clear Cooldown (asks first) | ⌘⇧K |
| Enable/Disable Account | ⌘⇧E |
"Use Rate-Limit Reset" spends one of the account's upstream reset credits or grants, and you can't get it back. "Clear Cooldown" only clears CLIProxyAPI's local routing cooldown and leaves upstream limits alone.
The six Store screenshots in metadata show Details, Limits, Overview, Forecast, Drain Rate and Chart Options using sample accounts and history. Overview demonstrates three accounts per provider. Every connected provider appears in the account views, Overview and Chart Options. Quota collection supports the built-in providers listed above and CPA-advertised quota plugins; providers without an exposed numeric quota remain visible with an explanation.
The extension records fresh quota observations whenever it loads quotas. The Record Quota History command also runs in the background about every 15 minutes while Raycast is running. Disable Collect Usage History in extension preferences to stop background sampling. Foreground refreshes still record observations. Raycast scheduling is approximate; sleep, closed Raycast and failed requests leave gaps.
History stays in local JSON files under the extension's support directory, separated by a hash of the server URL and management key. Observations contain opaque account identifiers, provider names, quota percentages, reset times and timestamps. They contain no emails, filenames, management keys or account tokens. Files older than 28 days are pruned on subsequent samples for that connection. Changing connection settings starts a separate history.
Hourly averages are derived from the stored observations, so they survive restarts without accumulating errors from averaging averages. For each account and quota window, the extension divides the decrease in remaining percentage points by observed time, splitting intervals across clock hours in proportion to their duration. Refills, changed reset windows, unknown reset times and gaps over 90 minutes are excluded. Each plotted hourly provider total needs at least 15 minutes of observations for every included account; otherwise the line has a gap. A flat measured interval is zero usage; missing history is not.
Forecasts use each account's time-weighted observed drain over the past seven days, not request or token counts. Each account needs at least one observed hour across three or more valid intervals before projection starts. The forecast applies that constant rate independently to each account, stops at zero, and assumes a full refill at its reported reset time. Windows with an explicit fixed duration repeat on that schedule; legacy Claude/Codex weekly and five-hour windows keep their known periods. Monthly or unknown-duration windows use only their next reported reset. Unknown reset times are projected without refills. No reset credits or grants are spent by the collector or forecast. Partial refreshes pause the projection.
This is a usage estimate, not a guarantee. Changes in workload, provider quota policies and rolling-window recovery can differ from a constant drain and full-refill model. One account's capacity is 100 percentage points, so three accounts have 300% capacity. A rate of 3%/h means three percentage points of one account's capacity per hour across those accounts.
Charts start in a learning state because the Management API does not provide the subscription quota history needed to backfill them. Request/token usage telemetry is a different measurement.
Charts use a full-width Detail view without a Grid card or selection outline. Press ⌘⇧P to open Chart Options, which combines graph, time range, provider and quota window in one searchable menu. Account views are available separately in Actions. Account Views sits at the bottom and includes icons. The canvas omits the duplicate time-range label, information footer and shortcut text. Raycast's Detail view has no main search input or top-right dropdown.
Chart shortcuts:
| Action | Shortcut |
|---|---|
| Open Forecast from account views / toggle Forecast and Drain Rate | ⌘T |
| Next provider | ⌘J |
| Next quota window | ⌘L |
| Toggle 24 hours / seven days in Drain Rate | ⌘H |
| Reload quotas and history from a chart | ⌘R |
| Open Chart Options | ⌘⇧P |
| Cycle all views | ⌘D |
The following dark-theme chart screenshots use sample history. Real history starts collecting after installation.



The extension discovers all credentials from CPA, including custom providers, instead of limiting the provider list. It prefers v8 credential and quota-proxy routes and falls back for older CPA servers. Every enabled credential with an auth index is considered for collection. Disabled accounts remain visible and are not polled.
Built-in readers follow the official Management Center: Antigravity quota groups, xAI weekly/monthly billing, Meta subscription windows, Kimi coding usage on both official domains, and Devin daily/weekly quota, alongside the existing Codex and Claude readers. A CPA-selected quota plugin takes precedence; custom providers use CPA's shared quota interface. Percentage-based windows enter local history and charts automatically, including model-specific groups. Plugin balance-only metrics appear as values rather than invented percentages.
Some API-key plans and providers do not expose a numeric usage limit. Those accounts remain visible and show that quota is not reported. The extension never runs inference to test a quota. Meta's DCA token and credential contents needed for project/domain resolution stay request-local and are not written to history or quota caches.
Accounts, quota windows, cooldowns and rate-limit resets
Record a quota sample for hourly usage and seven-day forecasts