Claude Code 2.1 saw significant improvements in August, focusing on cross-session collaboration, deployment flexibility, and user control over output and costs.
Collaboration and Communication
- Cross-session messaging: Sessions can now message each other by name using SendMessage and @mentions, with automatic discovery on macOS and Linux, and now Windows support too.
- Idle notifications: Ask another Claude Code session to send a one-time notice when it becomes idle, available on macOS and Linux.
- Effort syncing: Effort picks made on phone or claude.ai/code now sync in real time across terminal and Desktop/VS Code sessions.
Deployment and Configuration
- Self-hosted environments: Run Claude Code sessions on your own machines or containers via the self-hosted-runner command on Team and Enterprise plans.
- Multi-deployment support: Code review, cross-session messaging, and the /radio command now work on Bedrock, Vertex AI, and Foundry deployments.
- Plugin marketplace sources: Local IDEs can now print the plugin directory dynamically without restart, with GitLab support for nested subgroups.
- Model picker customization: The modelPicker setting lets you customize the model selector with an ordered, labeled list of models.
User Experience and Control
- Focus view: Hide tool activity behind expandable summaries with a live indicator, toggled via Ctrl+Alt+F.
- Concise output style: A new output style that delivers results first without preamble or narration.
- Cost optimization command: The /claude-api cost-optimize command profiles your Claude API spending and suggests reductions through caching, batching, and model selection.
Several other features were also released this month but are not listed here.