nodecode-ai-usage

AI usage: how much of each AI subscription is left, read from the plan itself. A Nodecode rewrite of lvabarajithan/hermes-ai-usage-tracker by lvabarajithan.

One GET per provider, to the provider that owns the key: GitHub Copilot's premium requests, chat and completions; Z.AI / GLM's coding-plan session, weekly and tool quotas; Kimi's coding plan; MiniMax's 5-hour and weekly windows; DeepSeek's balance; OpenCode Go's rolling, weekly and monthly windows; OpenCode Zen's credits. Read-only, and nothing goes anywhere else.

Get it

Tell the agent "add the ai-usage cell", or run nodecode add nc://ai-usage.

Set it up

Each provider's key is read from an environment variable, and the section names which variable for each:

Provider	Member	Usual variable
GitHub Copilot	`copilot_token_env`	`COPILOT_GITHUB_TOKEN`
Z.AI / GLM	`zai_key_env`	`GLM_API_KEY`
Kimi coding plan	`kimi_key_env`	`KIMI_API_KEY`
MiniMax coding plan	`minimax_key_env`	`MINIMAX_API_KEY`
DeepSeek	`deepseek_key_env`	`DEEPSEEK_API_KEY`
OpenCode Go	`opencode_go_key_env`	`OPENCODE_GO_API_KEY`
OpenCode Zen	`opencode_zen_key_env`	`OPENCODE_ZEN_API_KEY`

Export the ones you pay for. With none set the cell refuses at start, naming the variables it looked for; with one, it installs and reports the plans it has a key for, saying what to set for the ones it does not. cache_seconds (60) bounds how long an answer is reused and timeout_seconds (12) bounds one probe.

Use it

Ask, and the model calls the tool:

Tool	What it does
`ai_usage`	every provider whose key this home has, or one named: each quota window's used and left percentage, its reset and the plan's own detail

By hand, from the shell: /quotas prints the same report, /quotas refresh asks the plans again, /quotas zai reads one provider. From eval:

(ai-usage:report)                          ; the text /quotas shows
(ai-usage:report :provider "copilot" :refresh t)
(ai-usage:windows "zai")                   ; :windows, :plan, :reason as a plist

Provider ids are the original plugin's: copilot, zai, kimi-coding, minimax, deepseek, opencode-go, opencode-zen.

A plan that rejects the key, an endpoint its tier does not offer, a provider with no quota API and a missing key are each one line of the report, not a failure. A refusal reads ERROR: AI-USAGE-ERROR and says what to set.

The section, documented, is in config.example.jsonc.