NTM: Tmux Manager for Coordinating AI Coding Agents

What changed

NTM, a new tool developed in Go, aims to enhance the management of multiple AI coding agents within the tmux terminal multiplexer. It transforms tmux into a local control plane for multi-agent software development, offering features beyond basic pane management. The tool facilitates session orchestration, graph-aware work triage, safety policy enforcement, agent mail coordination, durable state capture, and exposes machine-readable robot surfaces through a local REST/WebSocket API. The latest release, v1.21.0, introduces a TUI command palette for managing these agents. NTM supports spawning and coordinating agents such as Claude Code, Codex, and Gemini (via Grok Build CLI). It also integrates with other tools like br, bv, Agent Mail, cass, dcg, and pt for advanced workflows. Key functionalities include spawning labeled multi-agent sessions, sending work and interrupts across panes, triaging next steps, coordinating agents via mail and file reservations, protecting dangerous operations with policies and approvals, and capturing state through checkpoints and audit trails. NTM also provides automation surfaces via JSON, REST API, SSE/WebSocket streams, and OpenAPI.

Agent Integration Details:

• Claude Code: Can be spawned with specific models (e.g., opus, sonnet, haiku) or use default settings. Configuration can be managed via ~/.config/ntm/config.toml.
• Grok Build (Gemini CLI): NTM recognizes the official xAI Grok Build CLI. It supports configuration, model selection, and effort arguments. Phase one includes basic launch and status projections, with advanced features like authenticated TUI readiness and automated prompt delivery planned for future development.
• Oh My Pi (omp): NTM operates omp as a built-in agent type. It supports specifying models and thinking levels, and can adopt existing omp panes. NTM derives its interactions from omp's live TUI, reading status, handling delivery, and managing interrupts.

Why it matters for builders

NTM provides developers with a structured environment for managing concurrent AI coding agents, mitigating the common difficulties of sustaining parallel workflows. By offering features like durable coordination, intelligent work triage, and robust safety policies, NTM allows builders to maintain better control over complex AI-assisted development tasks. This leads to more organized, efficient, and secure use of AI agents in their projects.

Practical impact

Developers can install NTM using a provided curl command for an easy setup. After enabling shell integration, they can verify dependencies and scaffold new projects. The tool allows for launching mixed swarms of agents (e.g., ntm spawn api --cc=2 --cod=1 --agy=1), opening live operator surfaces like dashboards and palettes, and dispatching work to agents. For projects utilizing br/bv, NTM can display the work graph. Saving recoverable checkpoints and serving local APIs for external access are also key capabilities. Builders can experiment with orchestrating multi-agent sessions, managing agent configurations, and leveraging NTM's automation surfaces for integration into their existing toolchains.

Caveats and source limits

The source material describes NTM as a "fresh release" with "4 AI signals, 5 developer signals" and provides a GitHub star count of 414 and 61 forks. Specific performance benchmarks or independent validation of its coordination and safety features are not detailed in the provided excerpts. While Grok Build integration is mentioned, some advanced features for this agent type are noted as not yet claimed in phase one. The exact pricing or licensing model for NTM itself is not specified, though it is a Go binary. The source does not provide details on the exact release date beyond "fresh release" and a published timestamp of "2026-08-05T16:59:42.000Z" which appears to be a future date and likely an artifact of the data source. The full capabilities of the REST/WebSocket API and OpenAPI specifications are not exhaustively detailed.