### Issue: - There is currently no mechanism to enforce a budget or abort when spending exceeds a threshold. - A single agent stuck in a tool loop can consume thousands of tokens per turn. - Users running against paid APIs (e.g., Anthropic, OpenAI) have no safeguards against runaway costs. - Implementing token budget enforcement would be high-value and low-effort. ### Proposed implementation: - Allow passing a max_token_budget configuration value.
Issue:
Proposed implementation: