Track quotas, costs, and accounts across 10+ AI providers.
Built-in proxies for Claude Code, Codex, OpenCode and Claude Science — 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
| 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 |
| Claude Science Proxy | Launch local Claude Science without a Claude subscription and route its inference to any third-party model; local virtual login and an isolated sandbox, with optional adoption so the double-clicked desktop app is login-free too — never touching your real credentials |
| 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 |
|
 |
| Dashboard | Provider & Multi-account Monitoring |
 |
 |
| Claude Code Proxy · Node Management | Claude Code Proxy · Configuration |
 |
 |
| Codex Proxy · Nodes & Subscriptions | OpenCode Proxy · Nodes & Stats |
 |
|
| Claude Science Proxy · Login-free Local Science | |
 |
 |
| Usage Stats (Claude, Codex & OpenCode) | Menu Bar |
 |
|
| Call Analytics · MCP / Skill / Tool Usage | |
Download .dmg or .zip from the Releases page.
Universal Binary — runs natively on both Apple Silicon and Intel Macs (macOS 14+).
AIUsage ships four independent proxies — for Claude Code, Codex (Codex CLI), OpenCode and Claude Science — each with node management, usage logging and a unified switcher.
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.
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.
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).
Launch a local Claude Science instance without a Claude subscription, and route its inference to a third-party model of your choice through the local proxy — while keeping tool calls, Skills, MCP and code execution intact. For personal study and research only; use at your own risk.
| Capability | What it does |
|---|---|
| Local virtual login | Forges a self-made virtual OAuth credential in an isolated data-dir to pass the login gate — zero Anthropic contact, never touching your real ~/.claude-science |
| Inference via third-party | Points ANTHROPIC_BASE_URL at the reused local QuotaServer, strips inbound OAuth, injects your third-party key, and maps opus/sonnet/haiku tiers to the node's real models |
| Isolated sandbox | Separate HOME / port (14410) / data-dir / keychain, zero impact on the real instance; one click opens the logged-in page in your browser |
| Adopt the real instance (optional) | An 8765 reverse proxy plus a decoupled internal daemon (14411) makes the double-clicked desktop app login-free too; session bootstrap tolerates current daemon response/cookie formats and reports redacted diagnostics if upstream auth changes |
| Shared node pool | Reuses the Claude-family nodes from the Claude Code proxy; hot-swap the upstream at runtime, transparent to Science |
Quick start: Prepare an upstream node on the Claude Code proxy page first → Open AIUsage → Claude Science Proxy → pick a node → One-click start; your browser opens the logged-in Science automatically. See docs/CLAUDE_SCIENCE_INTEGRATION.md for the technical design.
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 local port once; switching upstreams is in-process and CLI-transparent |
| LAN access (optional) | Bind the global proxy to 0.0.0.0 so other devices on your LAN can reach it via your machine's IP (off by default) |
| 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 |
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.
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.mdreads. Rule-hit counts are not tracked: rules are injected as context, not discrete calls.
Inspired by CodexBar and Quotio.
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.
If AIUsage helps you, consider buying the author a coffee. Your support helps keep the project maintained and improved.
