@deep-tui/plugin-budget

Adds per-run step, elapsed-time, token, and calculated-cost limits through the agent lifecycle extension. A reached limit stops before the next model request with the normal limit-reached run status, preserving the conversation and usage accumulated so far.

{
  "use": "@deep-tui/plugin-budget",
  "config": {
    "maxSteps": 100,
    "maxDurationMs": 1800000,
    "maxTotalTokens": 1000000,
    "maxCostUsd": 2
  }
}

maxSteps is opt-in; runs have no model-step limit when it is omitted. maxDurationMs defaults to 30 minutes. Token and cost limits are also opt-in because not every provider reports them. Usage limits are necessarily soft by one model response: the plugin can account for a response only after the provider reports its usage. Time is checked between model steps; provider-level request timeouts remain a separate concern.

Cost limits consume the provider's calculated charge, so DeepSeek requests use the peak or off-peak tariff active when each request starts. During DeepSeek's 01:00–04:00 and 06:00–10:00 UTC peak windows, the budget status displays DEEPSEEK PEAK and /budget status notes that the active rates are twice the off-peak rates.

Use /budget, /budget status, /budget on, or /budget off to inspect or temporarily disable the policy for the current Deep TUI process.