Juggle rate limits across all your Codex and Claude Code accounts. See every token you burn.

_{macOS · Bun · a Rubric Labs project · not affiliated with OpenAI or Anthropic}

Install

bun add -g tokenmaxx

tokenmaxx login codex     # sign in; your existing sessions stay untouched
tokenmaxx login claude
tokenmaxx install         # route native codex & claude through tokenmaxx

codex                     # use your clients as before
claude                    # tokenmaxx injects the account, per request

What it does

You run a fleet of coding agents, and they burn through the five-hour or weekly limit on whichever subscription you're using:

tokenmaxx runs a loopback proxy that reads the active account on every request and injects its credential, so a switch lands on the very next request, even mid-turn, with your clients running unmodified. Credentials stay in the macOS Keychain; nothing but opaque references touches disk.

The dashboard

Run tokenmaxx. Accounts shows every account and its live rate-limit windows, colored by pressure, with plan tier and reset countdowns inline. Rows sort by pressure; the ● marks where traffic is going right now.

Analytics is combined token throughput across all accounts and both providers, with the ≈ API-list-price value you're pulling from your flat subscriptions. Tokens are metered as responses stream by, never buffered, so every number is cross-checkable against your clients' own session logs. Press m for the full pricing breakdown per model.

Settings holds the master on/off per provider, then auto-rotation, the switch threshold, and cooldown, applied live.

Auto-rotation

Turn it on and tokenmaxx steps off an account the moment its fullest window crosses your threshold, onto the one with the most headroom. If an account still hits its hard limit mid-flight, the proxy rotates and retries that request before your client ever sees the 429. Off by default; enabling it is your confirmation that your provider permits this use.

tokenmaxx auto both on --threshold 90    # or: codex | claude … off

How it works

A single loopback proxy on 127.0.0.1:8459, and the clients you already use.

• Account per request. The proxy reads which account is active for each request and injects its credential. Switching lands on the next request.
• Pressure read for free. Both providers report rate-limit state on every response; the proxy reads it as traffic streams by, so it always knows how full the active account is, with zero extra requests.
• Harness-agnostic. Anything speaking the Anthropic or OpenAI Responses API can point at /anthropic or /openai and get the same juggling.

Commands

tokenmaxx                                  live dashboard
tokenmaxx login <codex|claude>             sign in; isolated, idempotent
tokenmaxx install                          route native codex & claude
tokenmaxx uninstall                        restore native config
tokenmaxx switch <codex|claude> <email>    make an account active
tokenmaxx auto <both|codex|claude> <on|off> [--threshold N]
tokenmaxx list | status | refresh | doctor

Env: TOKENMAXX_HOME, TOKENMAXX_PROXY_PORT, TOKENMAXX_THEME.

Not affiliated

An independent Rubric Labs project, not an official product of, affiliated with, or endorsed by OpenAI or Anthropic. Use only accounts you own, and only where the relevant terms permit account automation. Inspired by codex-account-switcher.

_{Built by Rubric Labs · go nuts}