ChatMLchatml
AI Agent

Budget Controls

Manage costs with spending limits, turn caps, thinking budgets, and spend analytics.

Budget controls help you manage the cost and scope of AI operations.

Available Controls

ControlDescription
Cost limitMaximum USD budget per conversation (maxBudgetUsd)
Turn limitMaximum number of agent turns per conversation (maxTurns)
Thinking budgetMaximum tokens for extended thinking (maxThinkingTokens)

Real-Time Tracking

During streaming, ChatML displays:

  • Current cost for the conversation
  • Number of turns used
  • Token usage (input, output, cache read, cache creation)
  • Context window utilization percentage

Setting Limits

Configure budget controls in Settings > AI:

  1. Set your desired limits
  2. Limits apply per conversation
  3. When a limit is reached, the agent stops and notifies you

Spend Analytics

The Mission Control Dashboard includes a Spend Tracker that visualizes your API costs:

  • 14-day rolling history — See your spending trend over two weeks
  • Per-model breakdown — Understand cost distribution across Claude Opus, Sonnet, and Haiku
  • Hourly breakdown — Identify peak usage periods
  • Per-session costs — See how much each active session is consuming

Cost Management Tips

  • Start with conservative limits when exploring unfamiliar tasks
  • Increase limits for complex, multi-step work
  • Use turn limits to prevent runaway agent loops
  • Monitor context window usage — high utilization can affect response quality
  • Watch for sub-agent usage, which multiplies cost since parallel agents each consume tokens

On this page