choosing a terminal for Claude Code (and every other CLI agent)

Claude Code runs in any terminal — so "best terminal" guides that pretend otherwise are selling something. What's true is that coding agents change the job description: turns run for minutes, conversations carry state, and you're suddenly running four projects at once. Full disclosure up front: this guide is written by the korren project, korren is in the table, and we'll tell you when it isn't the right pick.

##what actually matters with agents in the loop

• Sessions survive restarts. An agent conversation is state. If your terminal forgets what was running, every reboot costs you context reconstruction. claude --resume exists — does your terminal use it?
• You hear about turn completion. Agent turns take minutes. You'll be elsewhere. A terminal that notifies when a run finishes — or needs input — while unfocused turns dead waiting time into parallel work.
• Parallel projects have somewhere to live. Agents multiply how much you juggle. Tabs alone stop scaling; you want workspace-level separation with per-project state.
• Output is reviewable. Agents produce a lot of scrollback. Command blocks (jump between runs, see what each did) beat one endless stream.
• Streaming throughput. Agents print fast and constantly; GPU rendering keeps the UI honest under load.
• Your code stays yours. The agent already reads your repo — your terminal doesn't also need an account and a telemetry pipeline.

##seven terminals, compared honestly

Reading the table fairly: "~" isn't a failing grade — it means the capability exists but needs configuration or plugins. Minimalist terminals are minimal on purpose.

##the honest recommendations

you run agents across several projects, locally, on linux or macos → korren

This is the case korren is built for. A workspace per project, the agent in a pane beside your editor and git tool, focus-aware notifications when a turn ends, and — the part nothing else does — session restore that brings back a Claude Code pane mid-conversation by resuming its exact session id. No account, no telemetry, agent never bolted into the core.

your team wants an agent orchestration platform → Warp

Warp has gone all-in on agentic development: multi-agent management, cloud hand-off, team features, and first-class Claude Code integration. If you want the terminal itself to be the agent platform — and the account and defaults that come with a platform — it's the deepest option, and it went open source in 2026.

you want the fastest, quietest possible glass → Ghostty or Alacritty

If your model is "the terminal should disappear," Ghostty (native, polished, fast) or Alacritty + tmux (maximum minimalism) are excellent. You'll assemble persistence and notifications yourself.

you're macos-only and want native depth → iTerm2

Two decades of macOS-native features, mature session restore, and tmux control mode. It predates the agent era but handles it fine.

you want a scriptable power-tool → kitty or WezTerm

Both give you programmable control (kitty's remote control, WezTerm's lua) to build your own agent workflow. More assembly, more possibility.

##frequently asked