Token Budget
A cap on how many tokens an agent may spend on a task before it has to stop or economize.
Think of it like
A daily allowance — once it’s gone, you either wrap up or do the rest more cheaply.
Example
A run is given a 200,000-token budget, and as it approaches the limit the agent starts summarizing instead of keeping full history.
How it actually works
A token budget is both a cost control and a forcing function: knowing it’s finite pushes the agent (and its designer) to spend context wisely. It ties directly to stop conditions — hit the budget and the run ends or degrades gracefully. The design question is what happens at the limit: hard stop, summarize-and-continue, or ask a human, each with a different tradeoff between completeness and cost.
For product teams
Budgets keep agent costs predictable and force hard choices about what’s worth spending context on.
For engineers
A per-run token ceiling that triggers summarization, degradation, or termination as it’s approached.
Related
- Cost Tracking — How you know what budget to set.
- Stop Condition — A budget-driven stop.
- Context window — The scarce space budgets are spent on.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome