Skip to content

Managed Agents

Released (v0.17.0)Available in stable release v0.17.0.

While NMSh is not an AI-first terminal, it offers powerful integrations with managed coding agents for those who want them. These integrations are completely optional and do not phone home by default.

NMSh provides a built-in assistant to help you navigate its commands and settings.

  • Type /btw (or /ask) followed by a question to get local guidance mapped to typed NMSh actions.
  • By default, this uses a deterministic local engine.
  • You can opt-in to using a local model (Qwen Local Understanding) through /llm. Models are never downloaded without explicit Yes confirmation.
Unreleased (v0.18.0)Implemented on the development line and slated for release in v0.18.0.PR #345Issue #305

If you use Claude Code, NMSh can display its session context in your prompt or Context Rail.

NMSh uses Claude Code’s official status-line interface, never terminal scraping.

To set it up:

  1. Run nmsh agent-status setup. This shows the exact change to Claude Code’s settings.json.
  2. Apply the change.
  3. NMSh will keep an allowlisted, private record per NMSh session.
  4. Modules can now show model, effort, context-window use, and 5-hour/7-day limits.

Privacy: Prompts, transcripts, and terminal output are never read by NMSh’s agent status bridge. The data is display-only and is not kept in history.