The native desktop for your agent team.
Run Claude Code, Codex and OpenCode agents in real terminals, see at a glance which one needs you, and read their reports and documents beside the work.
Website · Install · User guide · Architecture · Security · Contribute
Illustration with example data.
Agent Interface and Agent Backbone work together. Backbone turns agents across CLIs and projects into one team, with names, safe messages, live state and continuity. Agent Interface is where you work with that team.
- Know who needs you. Active agents show their state, such as working, idle or waiting on you. Click one to go to the tab that shows its session.
- Real terminals. Panes are drawn natively with GPUI and Ghostty's terminal core. tmux keeps every process running, and you can attach the sessions you already use.
- The team beside the work. Open Team, Activity, Usage, Skills, Policies and Tags views in a pane, next to the agents they describe.
- Reports, answers and documents. Reports & updates collects progress reports, saved answers and Documents. Reply to the author from the record you are reading.
- Workspaces that come back. Tabs and split panes are saved in workspaces. Jump (⌘K) finds tabs, panes, agents and commands.
Without Backbone, Agent Interface is still a native workspace for tmux.
flowchart LR
you([You]) --> app[Agent Interface]
app -- terminals --> tmux[tmux]
app -- "team, reports, documents" --> backbone[Agent Backbone]
backbone -- "messages and state" --> agents["Agents: Claude Code, Codex, OpenCode"]
tmux --- agents
You need macOS or Linux and tmux 3.4 or newer (brew install tmux on macOS).
On Linux, install the runtime libraries first.
1. Install Agent Interface and Agent Backbone. The installer downloads the compiled app from the latest release, checks its checksum, and installs the latest compatible published Backbone. It supplies uv and Python when needed; no Rust or manual Python installation is required.
curl -fsSL https://raw.githubusercontent.com/eandualem/agent-interface/main/install.sh | sh2. Open Agent Interface. Native onboarding creates your first profile, signs you in to an agent tool and starts your first agent, its guide. Existing Backbone configuration is preserved; see connection guidance if you manage that installation independently.
The installation guide covers package architectures, manual downloads, Windows 11 / WSL setup, updates and removal.
- Press ⌘T (Ctrl+Shift+T on Linux), then choose Open terminal or pick a running tmux session to attach it.
- Split the pane with ⌘D, or ⌘⇧D to split below (Ctrl+Shift+D and Ctrl+Shift+E on Linux).
- In an empty pane, choose a view such as Team or Reports & updates.
- Once an agent's session is in a tab, click the agent in the sidebar to go there.
The user guide covers every view, shortcut and setting.
- User guide: workspaces, sessions, agent views, reports, shortcuts and libraries.
- Installation: packages, prerequisites, Backbone connection, updates and removal.
- Architecture: processes, crate boundaries, data flow and dependencies.
- Security: the security model, how the app is tested, open security work and how to report a vulnerability.
- Design: appearance, components and design tokens.
- Benchmarks: repeatable measurements on isolated tmux servers.
- Contributing: building from source, tests and pull requests.
Agent Interface runs on macOS and Linux (X11, or XWayland on Wayland desktops). Windows 11 runs the Linux package through WSL 2 with WSLg. Sessions run on the same computer as the app. Rich HTML Documents render on macOS and Linux, including Windows through WSLg.
Not yet implemented: terminal input methods, key-release events and selection across history pages.
Agent Interface reimplements tmux-workspaces in Rust and opens its saved schema-v2 libraries. It needs no Python runtime.
MIT. See LICENSE.