Upstream project
Unterm vs WezTerm
Unterm is built on WezTerm's renderer and mux — then wraps it in an agent-control surface WezTerm doesn't ship.
WezTerm is an excellent, mature GPU terminal: Lua config, a battle-tested multiplexer, SSH domains, and cross-platform builds, all MIT-licensed. Unterm inherits that foundation and answers a different question — how does an AI agent operate a terminal? — by exposing the whole app over MCP and a CLI.
| Capability | Unterm | WezTerm |
|---|---|---|
| Driveable from outside via MCP | ✓ | ✗ |
| Scriptable CLI for every action | ✓ | ~ pane cli |
| Built-in AI coding-agent launcher | ✓ ×5 | ✗ |
| Agents auto-wire to the terminal's MCP | ✓ | ✗ |
| Identity profiles (creds per window) | ✓ | ✗ |
| Multi-instance orchestration | ✓ | ✗ |
| Session recording → markdown | ✓ md | ✗ |
| Screenshot + file upload from MCP | ✓ | ✗ |
| Built-in proxy management | ✓ | ✗ |
| Local-first, no cloud / no login | ✓ | ✓ |
| AI baked inside the terminal | ✓ agent-driven | ✗ |
| GPU rendering | ✓ | ✓ |
| macOS + Linux + Windows | ✓ | ✓ |
| Open-source client | ✓ MIT | ✓ MIT |
| Native UI languages | ✓ ×9 | en |
| Price | $0 | $0 |
✓ yes · ✗ no · ~ partial · ×N count. The ✗ by design on "AI inside" is the whole thesis: Unterm keeps the AI outside and exposes the terminal as the surface it grips.
Why teams choose Unterm
- ✓ An MCP server built in. Any LLM agent can list/create/split panes, run commands, read the screen and scrollback, and capture screenshots — WezTerm has a control CLI for panes, but nothing an external agent speaks natively.
- ✓ One-click AI agents + auto-wiring. Install and sign in to Claude Code, Codex, Gemini CLI, OpenCode, or Aider in a click, and Unterm registers itself into each client's global MCP config automatically.
- ✓ Agent-workflow surface. Identity profiles (per-window secrets from the OS keychain), multi-instance windows, redacted Markdown session recording, scrollback capture, and probe-gated proxy — none of which live in stock WezTerm.
Where WezTerm is stronger
WezTerm is the mature upstream with a deeper, more configurable terminal surface (Lua scripting, extensive keybinding and multiplexer features) and a longer track record. If you don't want any AI/agent layer, WezTerm is the leaner, more customizable pure terminal — and Unterm tracks it, so you're not giving up its rendering.
Bottom line
Want a fast, scriptable GPU terminal and nothing else? Use WezTerm. Want that same engine turned into something an AI agent can actually drive — install agents, run them, read results — without gluing together scripts? That's exactly what Unterm adds on top.
MIT-licensed · macOS / Linux / Windows · no account, no login.