- Keep unnecessary context to a minimum in any conversations, tasks, and plans.
- Avoid hiring additional agents too fast too quickly.
- Check your lead and MCP permissions. If they are too loose, especially during the build phase, your agents can use tokens fast!
- Disable unused MCPs. They can use up tokens with unnecessary context.
- Choose the right model for the job. Experiment with cost vs intelligence tradeoffs.
- Try asking your agent to kill unnecessary running processes.
- Hit the ‘Reset chat’ button. This clears out stale context and kills running processes. Find a good moment to do this since you’ll lose your recent chat context.
- Reset or upgrade your sandbox. If this is available, a message will appear in your AI Business sidebar.
- Try resetting your chat. This clears out old context that may be corrupted.
- Switch to a different model from a different provider. Sometimes, model providers can experience downtime.