@tokenmix/opencode-plugin

Adds three tokenmix.ai account tools to OpenCode so the agent can browse models, open the top-up page, and hint about balance from inside a coding session.

Install

In your opencode.json:

{
  "plugin": ["@tokenmix/opencode-plugin"]
}

OpenCode auto-installs the npm package on first run.

What the plugin gives the agent

| Tool | What it does | |---|---| | tokenmix_models | List all tokenmix-served models and prices, optionally filtered by type (chat / image / video / audio / embedding) | | tokenmix_topup | Opens tokenmix.ai/dashboard/credits in the user's default browser | | tokenmix_balance | Returns your live balance — available, balance, gift credit, total spent — via the API-key-authenticated GET /v1/wallet, plus a top-up link | | tokenmix_usage | Reports token usage (input / output / total) for the current OpenCode session |

The agent decides when to call these — for example, when the user asks "show me what models I can use" or "I'm running low, top me up."

Low-balance alerts

When the assistant finishes a turn, the plugin checks your wallet (throttled to once a minute) and shows a one-time toast if your available balance drops below a threshold — so you're warned before a request fails. Set the threshold (USD, default 1) in opencode.json:

{
  "plugin": [["@tokenmix/opencode-plugin", { "lowBalanceThreshold": 2 }]]
}

Every wallet/TUI call is wrapped, so a network or headless-server failure never disrupts your session.

Companion CLI

Want zero-config setup of OpenCode + tokenmix? Try npx tokenmix opencode — installs OpenCode, signs you in by browser, writes the config, and launches.

Roadmap

• ~~v0.2 — Live balance via tokenmix /v1/wallet~~ ✅ shipped
• ~~v0.3 — per-session token usage (tokenmix_usage) + low-balance toast~~ ✅ shipped
• v0.4 — Optional cache_control injection for anthropic-routed requests (cheaper repeated context)
• Future — localized output (中文 / 日本語 / 한국어 / Español / Français), matching the CLI

License

MIT