Skip to content
Resource archive · retained for referenceExplore design skills →

Copilot CLI session limits: a per-task budget checklist

A practical guide to setting, observing, and reviewing AI credit limits for Copilot CLI and SDK agent sessions.

Copilot CLI and the Copilot SDK can now stop a session after a configured AI credit budget. Use that as a per-task control for unattended agent work, not as a replacement for permissions or monthly spending limits.

Scope One Task First

Session limits cover model calls, subagents, and background work across one session. Update to Copilot CLI 1.0.66+ or SDK 1.0.5+, then name the repository, task, and expected artifact before choosing a number.

Choose The Right Entry Point

  • Interactive CLI: `/limits set max-ai-credits NUMBER`.
  • Noninteractive CLI: `copilot -p "YOUR PROMPT" --max-ai-credits NUMBER`.
  • SDK: pass `sessionLimits.maxAiCredits` when creating or resuming the session.

Expect Soft-cap Overshoot

Usage is checked after a response returns, so an in-flight response can push actual usage above the configured number. Treat the value as a stop checkpoint, not a hard currency ceiling.

Make Exhaustion Observable

Preserve the session ID, partial artifact, test output, configured limit, and used credits. SDK hosts should subscribe to usage and exhaustion events, then require an explicit continue, add-credits, or stop decision.

Keep Monthly Budgets Separate

Per-session limits complement user and organization budgets; they do not replace them. Repeated exhaustion should trigger a review of task scope, context, model, tools, and retries instead of an automatic budget reset.

Know When To Skip

Skip this as the primary control when you need organization-wide billing governance, run an unsupported client, or have not yet defined permissions and rollback. The feature is in public preview, and Copilot CLI uses its own GitHub license terms.

Sources

Copilot CLI AI Credit Session Limit Checklist