Providers — UsageNow Docs

A provider is an AI coding tool UsageNow can report on. Each one exposes different information, so UsageNow shows only what a provider actually reports.

Available#

CodexUsage limits, reset times, plan, and local activity.Claude CodeLocal activity and plan, plus experimental usage limits.Gemini CLIActivity by model. No usage limits.AntigravityUsage limits from the IDE or compatible official CLI.KiroMonthly credit limit, plan, and today's credits.WarpToday's agent requests, credits, and models.OpenCodeToday's tokens, requests, models, and estimated cost.QoderCredits spent today and the answers they paid for.Qwen CodeToday's tokens, requests, models, and estimated cost.ClineToday's tokens, requests, models, and the cost Cline records.Grok BuildToday's tokens, model calls, models, and the cost xAI reports.DeepSeek, Kimi, OpenRouterBalance and spending, with your own API key.Ollama CloudSession and weekly limits, with your own API key. Experimental.WindsurfDaily and weekly limits from the local cached plan.JetBrains AIExperimental local AI Assistant quota. Active-subscription validation pending.CursorPersonal plan allowance, with a web session. Experimental.GitHub CopilotAllowances and credits, with your GitHub token. Experimental.PerplexityWeb credits, with a web session. Experimental.ElevenLabsMonthly characters, with a restricted API key.Z.aiGlobal Coding Plan quota, with your API key. Experimental.

Explicit connections#

Keys, tokens, and web sessions are added only by you. They are checked before saving to Keychain; HTTP 429 cooldowns also apply to manual refresh. See Connect a provider. Cursor, Copilot, Perplexity, and Z.ai use experimental endpoints that can change; compatibility depends on the account and plan.

More than one account#

Codex, Claude Code, and the explicit-connection providers can show several accounts side by side — a personal and a work Codex, say, or two DeepSeek keys. Add them in Settings › Providers; see Accounts.

Turning a provider off#

Every provider that reads a tool on your Mac is on by default, and appears once the tool is installed. Providers that need a key, token, or web session start off. Turn any one on or off in Settings › Providers, and drag providers into the order the popover and widget should use.

When a provider is off, UsageNow doesn’t:

refresh the provider
access the provider’s local files
access the provider’s credentials
include the provider’s values in menu bar calculations
include the provider in widget data

UsageNow never deletes or changes the provider's own files or sign-in. For explicit connections, turning a provider off deletes the credentials you saved in UsageNow's Keychain, including added accounts. Reconnect if you turn it back on.

How data differs between providers#

Providers measure usage in their own ways, so their numbers aren’t directly comparable:

Windows differ. One provider may report a 5-hour and a weekly window; another may report only one, or a window for a single model family.
Sources differ. Codex reports limits through its own app-server; Claude Code through an experimental source; Antigravity through its IDE or official CLI; JetBrains AI and Windsurf through local cached quota; Kiro through the limits it writes to its own log; Ollama Cloud through an experimental endpoint. Gemini CLI, Warp, OpenCode, Qoder, Qwen Code, Cline, and Grok Build report no limits. DeepSeek, Kimi, and OpenRouter report money rather than a percentage.
Meters differ. Most tools count tokens. Kiro and Qoder count credits, and Warp counts both. The 30-day history charts each tool in its own unit.
Account counters are separate. Credits or characters from connected accounts are not today's local token activity. ElevenLabs reports characters; Perplexity reports web credits, not Sonar API usage. Missing or unlimited allowances do not get a made-up percentage.
Local activity is separate. Tokens and requests come from session files on your Mac and aren’t the same as quota consumption. Where a provider records the model, UsageNow also shows activity by model.

UsageNow never invents a provider's missing limit. The separate quota pace forecast is explicitly labelled as an estimate and requires sufficient fresh observations.