A new open-source terminal, Unterm, hit Hacker News today, pitching a pure-Rust solution designed explicitly for a hybrid workflow where both humans and AI agents drive the interface. Unlike Warp, which embeds AI in a closed cloud, or Ghostty, which stays out of the way, Unterm bets on the terminal being a controllable surface for external agents via the Model Context Protocol (MCP). The project, hosted on GitHub by zhitongblog, aims to provide a local, secure, and high-performance environment for running CLI agents like Claude Code, Codex, and Gemini.

MCP as the Universal Interface

At its core, Unterm exposes a local MCP server on 127.0.0.1, allowing any AI agent to interact with the terminal using line-delimited JSON-RPC. This surface includes 152 authenticated methods, enabling agents to spawn shells, execute commands, read pane states, and even capture screenshots without needing a cloud round trip. The architecture ensures that agents can grip the terminal through the API while keeping the human in the loop via a unified cockpit, effectively separating the agent's logic from the terminal's rendering engine.

The Agent Cockpit and Governance

Since version 0.55, Unterm introduced the "Agent Cockpit," a feature set that allows the terminal to monitor and manage agents running inside it. This includes live per-pane state tracking, a waiting-first Inbox for agent prompts, and a "Fleet" mode that runs multiple agents on the same task in isolated git worktrees. With v0.68, the system added governance features like capability leases, workspaces that cannot see each other, and a hash-chained audit trail, ensuring that agent actions are verifiable and contained.

Cross-Platform Performance and Features

The terminal is built on Unterm's native "next-core" engine, replacing the previous WezTerm kernel to achieve a 2.8x faster cold start and significantly reduced CPU usage during output floods. It supports GPU-accelerated rendering across macOS, Linux, and Windows using wgpu, with fallbacks for systems without dedicated graphics drivers. Key features include identity profiles for managing secrets via OS-native vaults, multi-instance discovery using NATO phonetic names, and a Web Settings UI for modern configuration management.

Key Takeaways

  • Unterm uses MCP to allow external AI agents to control the terminal via JSON-RPC.
  • The "Agent Cockpit" manages fleets of agents in isolated git worktrees.
  • v0.68 adds governance features like capability leases and hash-chained audits.
  • Built on a pure-Rust "next-core" engine for cross-platform GPU acceleration.

The Bottom Line

Unterm is a necessary pivot for the agent era, proving that terminals must become governed, auditable surfaces rather than just passive text displays.