gilvt
Short answer: a normal terminal displays text. An agent terminal adds structured awareness across Claude Code, Codex and other long-running sessions, while keeping each agent in its native command-line interface.
Why a normal terminal becomes difficult with coding agents
A terminal window works well when one person drives one foreground command. Coding agents change that pattern. A developer may start several sessions, move to another task, and return only when an agent needs approval or finishes a turn.
With ordinary tabs and splits, every session looks like another rectangle of text. You have to open each tab to answer basic questions:
- Is this agent still working, or did it stop ten minutes ago?
- Which session is waiting for an approval or an answer?
- What files changed during the last turn?
- Can this older session be resumed safely?
The problem is no longer terminal rendering. It is attention management across asynchronous workers.
What an agent terminal adds
| Layer | Normal terminal | Agent terminal |
|---|---|---|
| Processes | Shows shell output | Associates a process with an Agent session |
| Status | Requires reading the current screen | Tracks running, waiting, error and completed states |
| Attention | Manual tab checking | Ranks sessions that need the user |
| Review | Open separate Git and editor tools | Connects turns, files, commands and diffs |
| History | Scrollback and shell history | Finds and resumes durable Agent sessions |
How gilvt observes agents
gilvt does not replace Claude Code or Codex with a proprietary chat interface. The agents continue to run in their own TUIs. gilvt combines shell integration, Agent hooks, session records and process information into a normalized status model.
That model powers a sidebar, pane borders, notifications, per-turn timelines and artifact review. If an integration is unavailable, the terminal continues to work and the session falls back to a reduced mode.
An agent terminal is not an autonomous orchestrator
There are two distinct product models. An orchestrator often owns prompts, approvals and the execution graph. An agent terminal can take a less prescriptive role: it observes, organizes and navigates without answering on the user's behalf.
gilvt follows the second model. It does not type into agents or approve commands. The user keeps the original CLI behavior and configuration, while gaining a workspace for running many sessions.
How it relates to tmux
tmux solves a different problem: it keeps terminal processes alive in a server so clients can detach and reconnect. An agent terminal focuses on native UI and Agent semantics. They work well together, especially when tmux keeps remote work alive and gilvt provides the local macOS interface.
Read the full gilvt versus tmux comparison for the process and session-lifecycle differences.
Try it: download gilvt for macOS, open a pane, and run claude or codex. The session appears in the sidebar automatically.