oneport-account

The account & metering client shared by every paid Oneport tool. Login, token balance, redeem codes — and the call_managed_llm() function tools use to run a billed model call through the Oneport backend.

No tool ever holds a model key. Each holds only your revocable op_live_... token; the backend meters your balance and calls the model with the server-side key.

For users

# 1. Get a token (free signup tokens) at https://oneport.dev, then:
oneport-account login op_live_xxxxxxxxxxxx      # or: op account login ...
oneport-account balance                         # tokens remaining + usage by tool
oneport-account redeem LAUNCH2026               # redeem a free-token code
oneport-account logout

When a tool runs out of tokens it prints a buy link; top up on the website and keep going. Codes grant free tokens (great for launches, students, design partners).

For tool authors

from oneport_account import call_managed_llm, OutOfTokens, NotLoggedIn

try:
    result = call_managed_llm(system, user, tool="apidiff", action="classify")
    text, tokens, balance = result.text, result.tokens_used, result.balance
except NotLoggedIn:
    ...   # tell the user to `oneport-account login`
except OutOfTokens as e:
    ...   # show e.buy_url

call_managed_llm raises NotLoggedIn, OutOfTokens (with .buy_url), RateLimited, or AccountError — map these onto your tool's own taxonomy.

Configuration

The backend URL resolves as: $ONEPORT_API_URL → the value pinned at login → DEFAULT_API_URL in config.py (set this to your Supabase Edge Functions base URL once, e.g. https://<ref>.supabase.co/functions/v1).

The token is stored at ~/.oneport/credentials.json (0600 where supported).

Privacy

Deterministic work in every tool stays local. Only the LLM judgment call routes through the proxy so it can be metered — and only the already-computed prompt (never your repository) is sent. See each tool's PRIVACY notes.

MIT licensed.