AIUsage

One dashboard for all your AI subscriptions

Track quotas, costs, and accounts across 10+ AI providers.
Built-in proxies for Claude Code, Codex and OpenCode — any OpenAI-compatible model.

中文说明 · English

^{Sponsored by}

_{500+ AI models · Text, image, video & audio · Top models included · Pay-as-you-go}

---

Table of Contents

• Features
• Preview
• Install
• Proxies
• Call Analytics
• Acknowledgements
• Sponsor
• Support the Author
• License

Features

Feature	Description
12+ AI Providers	Codex, Copilot, Cursor, Antigravity, Kiro, Warp, Gemini CLI, Droid, Claude Code, OpenCode, Kimi, MiniMax — one dashboard
Multi-account	Multiple accounts per provider, independent refresh, one-click CLI switching
Usage Stats	Unified cost & token breakdown from Claude/Codex proxy archives, token-only non-proxy Codex sessions, and OpenCode's local session ledger — per-model trends, time-period analysis, source-aware aggregation
Claude Code Proxy	Use Claude Code with DeepSeek, GPT, Ollama or any OpenAI-compatible model; Anthropic passthrough for usage logging
Codex Proxy	Point Codex CLI at any OpenAI-compatible upstream; unified switcher across subscription accounts and API nodes, surgical config.toml merge
OpenCode Proxy	Switch OpenCode across upstreams via a managed opencode.json block — OpenAI-compatible, Anthropic and Responses protocols, per-node usage attribution, per-model pricing and optional request logging
Global Proxy	One fixed local endpoint per agent — hot-swap the active upstream node with zero CLI restart, per-node cost attribution, and automatic cross-track port arbitration
Unified API Providers	Configure one upstream (Base URL, format, key, model library/pricing) once and distribute it to Codex / Claude / OpenCode at once; linked nodes inherit from the master and sync on change, with per-field local overrides
Call Analytics	Count MCP / Skill / tool calls across Claude Code, Codex & OpenCode from local session logs — Top-N rankings, daily trend, and per-app zero-call ("zombie" skill/MCP) detection; read-only, zero instrumentation
Menu Bar	Multi-account status bar icons, quota/cost metrics, per-proxy node switchers, quick-glance popover, colored progress bars
Credential Vault	macOS Keychain storage for all managed credentials

Preview

Dashboard	Provider & Multi-account Monitoring
Claude Code Proxy · Node Management	Claude Code Proxy · Configuration
Codex Proxy · Nodes & Subscriptions	OpenCode Proxy · Nodes & Stats
Usage Stats (Claude, Codex & OpenCode)	Menu Bar
Call Analytics · MCP / Skill / Tool Usage

Install

Download .dmg or .zip from the Releases page.

Proxies

AIUsage ships three independent proxies — for Claude Code, Codex (Codex CLI) and OpenCode — each with node management, usage logging and a unified switcher.

Claude Code Proxy

Use Claude Code CLI with any OpenAI-compatible model, or transparently log Anthropic API usage.

Mode	What it does
OpenAI Proxy	Translates Claude API → OpenAI format. Works with DeepSeek, GPT, Azure, Ollama, etc.
Anthropic Passthrough	Forwards requests as-is, logs input/output/cache tokens for cost tracking

Quick start: Open AIUsage → Claude Code Proxy → New Node → Configure → Activate. Settings are written to ~/.claude/settings.json automatically.

Codex Proxy

Point the Codex CLI at any OpenAI-compatible upstream (Responses API), and switch between subscription accounts and API nodes from one place — they are mutually exclusive, so only one identity is ever active.

Capability	What it does
OpenAI-compatible upstream	Routes Codex CLI through any responses-compatible endpoint
Unified switcher	One toggle across subscription accounts (~/.codex/auth.json) and API nodes (config.toml)
Surgical config merge	Injects managed blocks into ~/.codex/config.toml while preserving your own settings; global fragment + per-node TOML override
cc-switch sync	One-click import of Codex providers from local cc-switch (upstream / model / key), preserving model_reasoning_effort, mcp_servers and other settings; deterministic-id dedup avoids duplicate nodes (symmetric with Claude)

Quick start: Open AIUsage → Codex Proxy → New Node (or pick a subscription account) → Configure → Activate. ~/.codex/config.toml is merged automatically. Already on cc-switch? Hit "Sync cc-switch" in the toolbar to import in one click.

OpenCode Proxy

Switch OpenCode between any number of upstreams without hand-editing opencode.json. AIUsage injects a managed provider block (and points the top-level model at it), then restores your original config on deactivate — the backup is the source of truth, so takeover is idempotent.

Capability	What it does
Multi-protocol	OpenAI-compatible (@ai-sdk/openai-compatible), Anthropic (@ai-sdk/anthropic) and OpenAI Responses (@ai-sdk/openai) — the npm package follows the node's protocol
Direct or proxy mode	Direct rewrites opencode.json to talk to the upstream; proxy mode points it at a local passthrough process for per-request logs (usage/cost still comes from opencode.db)
Per-node usage	Each node writes a distinct managed providerID, so the local OpenCode session ledger attributes tokens and models.dev-priced cost back to the right node
Model library & pricing	Per-node model list with independent per-model pricing (USD/CNY); pick the default model from the library and switch from the node card
cc-switch sync	One-click import of OpenCode providers from local cc-switch (upstream / models / key / pricing), deterministic-id dedup, configurable cc-switch directory

Quick start: Open AIUsage → OpenCode Proxy → New Node → Configure models & pricing → Activate. ~/.config/opencode/opencode.json is taken over automatically (requires OpenCode ≥ 1.2 for usage tracking).

Global Proxy

Instead of activating one node at a time, run a single fixed local endpoint per agent and hot-swap the active upstream behind it — the CLI never restarts and its config never changes.

Capability	What it does
Fixed endpoint	Each agent points at one stable 127.0.0.1 port once; switching upstreams is in-process and CLI-transparent
Hot-swap active node	Change the active node with zero restart; each request is rewritten to that node's real upstream model
Per-node attribution	Cost and usage are recorded against the real active node and model, not a generic global bucket; ports are arbitrated across all three tracks to avoid collisions

Unified API Providers

Configure an upstream once and reuse it everywhere. Under Providers → API Providers, define a provider (Base URL, API format, key, model library and pricing) and distribute it to any combination of the three proxies — each gets a linked node.

Capability	What it does
Configure once, distribute	One config feeds Codex / Claude / OpenCode; a compatibility matrix limits where each format can go
Inherit + local override	Linked nodes follow the master and sync on change; editing a shared field on one node turns it into a local override that no longer syncs
Safe lifecycle	Idempotent re-distribution (no duplicates), port deconfliction for new nodes, and cascade delete or unlink when removing a master

Usage and billing details for Claude Code and Codex are documented in docs/USAGE_AND_BILLING.md. OpenCode cost is read directly from its local session ledger (opencode.db), pre-priced per models.dev.

---

Call Analytics

See which MCP servers, skills and tools you actually use. AIUsage parses the local session logs of Claude Code, Codex and OpenCode — read-only, with zero instrumentation — and turns tool calls into usage insights: spot the MCP servers you rely on, and clean up "zombie" skills that were installed but never called.

Capability	What it does
Top-N rankings	MCP servers (folded by server), skills and built-in tools ranked by call count, with stable tie-breaking
Daily trend	Per-day call volume over the selected 7 / 30 / 90-day or all-time window
Zero-call detection	Flags installed-but-never-called skills and configured-but-unused MCP servers — scoped per app, so each tool's own cleanup candidates are actionable
Per-source scope	Filter by Claude Code / Codex / OpenCode or view all; skills and MCP are attributed to each tool by its own skill directories and config files

Codex skill calls are heuristic — Codex has no discrete skill-invocation event, so they are inferred from SKILL.md reads. Rule-hit counts are not tracked: rules are injected as context, not discrete calls.

---

Acknowledgements

Inspired by CodexBar and Quotio.

Sponsor

Sucloud — AI API aggregation platform with 500+ models.
Full modality coverage (text, image, video, audio) including Claude, GPT, Gemini and more.
RMB payment supported, no overseas card required.

Support the Author

If AIUsage helps you, consider buying the author a coffee. Your support helps keep the project maintained and improved.

Friendly Links

• Linux.do Community

License

Apache License 2.0