Tend, a new open-source utility released by Geekgineer, is addressing the orchestration crisis for power users running multiple Claude Code instances. The tool provides a single local web page to monitor, control, and review up to 25 concurrent Claude Code sessions, turning a chaotic wall of terminals into a unified control plane. Unlike wrappers that abstract the model, Tend operates as a lightweight listener for Claude Code's native hooks and status lines, ensuring that the underlying TUI remains unmodified.

Architecture and Performance

The core technical achievement of Tend is its ability to aggregate session state without introducing latency or AI costs. By binding exclusively to localhost (127.0.0.1:4977) and utilizing Server-Sent Events (SSE) alongside tmux control mode, Tend achieves a real-time view of session activity. It leverages a SQLite database for state persistence and requires only five runtime dependencies, avoiding the bloat of native modules. This architecture ensures that the tool remains transparent to the user's existing workflow while providing a high-throughput interface for session management.

Context-Aware Decision Making

Tend introduces a 'risk label' system that categorizes permission requests and plan approvals using rule-based logic rather than AI inference. This approach ensures that high-risk actions are never auto-approved and that users retain granular control over their codebase. The tool supports a 'Review' tab that mirrors the experience of reviewing a Pull Request, offering unified or split diffs with word-level highlighting. This allows developers to inspect changes across multiple sessions simultaneously, a significant improvement over the linear review process of traditional terminal-based workflows.

Integration and Compatibility

The tool is designed for macOS and Linux environments, requiring Node.js 24+ and tmux. It supports the adoption of existing sessions started outside of Tend, allowing users to 'adopt' background agents into the unified view. A notable feature is the 'Briefing' system, which generates session summaries using plain rules and event data, ensuring that no additional tokens are consumed by the orchestration layer itself. This 'zero-token' philosophy extends to the entire stack, making it a cost-neutral addition to the Claude Code ecosystem.

Key Takeaways

  • Tend aggregates up to 25 Claude Code sessions into a single browser interface without modifying the underlying TUI.
  • The tool uses rule-based risk labels and local SQLite storage to ensure zero AI token overhead for orchestration.
  • It supports macOS and Linux, requiring Node.js 24+ and tmux, with a focus on local-only privacy and security.
  • Developers can review changes across sessions using a PR-style interface with unified diffs and word-level marks.

The Bottom Line

Tend is a necessary evolution for the 'multi-agent' era of coding, finally providing the UI layer that terminal-based LLMs lack. It proves that effective orchestration doesn't require a proprietary cloud service or additional token spendβ€”it just requires a better way to look at your own terminal.