Your existing allowance. Visible learning usage.

Learning is additional agent work through the host’s model connection. It is not free inference and does not require a separate DontDumb AI account.

What consumes tokens

• Selecting a useful learning opportunity from context.
• Generating explanations, questions, maps and interactives.
• Assessing answers, providing hints and creating recaps.

Work sampling reduces forwarded context. It does not eliminate native host instructions, tool overhead or provider costs.

Read the counters

Learn settings show learning input/output tokens separately, call counts and session/all-session totals. These are learning counters, not the main coding agent’s total usage. Host-reported counts are used when available; missing usage may be estimated. Counters are not an exact invoice or currency cost.

Model selection

The learning model and content-builder preferences use available host models. A cheaper model is only possible where your existing host connection supports it. DontDumb does not silently introduce a new paid provider or API key.

Control consumption

Adjust learning frequency, pause a session, or turn learning off. Closing Learn hides the panel while observation continues. The current counters are informational and do not enforce a hard spending cap.

Shared limits

Subscription quotas, provider rate limits and local resources may be shared with coding. Independent learning avoids blocking on learner actions; it cannot guarantee zero throughput impact.