Warp: Kitty Keyboard Protocol Support and Customizable Agent Toolbelt

Kitty Keyboard Protocol Support

What It Is and Why It Matters

Warp now supports the Kitty Keyboard Protocol, a modern extension to how terminals communicate keyboard input to applications. Traditional terminal keyboard handling has a fundamental limitation: many key combinations share identical escape codes, making it impossible for terminal applications to tell them apart. Modifier+key combinations frequently collide, and applications have historically had no way to detect key release events — only presses.

The Kitty Keyboard Protocol resolves these issues with a structured, unambiguous input layer. With it, terminal applications running inside Warp gain several new capabilities:

• Full modifier key detection — support for shift, alt, ctrl, super, hyper, and meta, rather than the legacy subset
• Disambiguation of identical key codes — such as Esc vs. the start of an escape sequence, which previously mapped to the same bytes
• Press, repeat, and release events — essential for applications that respond differently to holding a key vs. tapping it
• Improved international keyboard and IME support — the protocol separates key codes from their associated text output, handling non-Latin input more robustly

The protocol has been adopted broadly across the terminal ecosystem: Kitty, Alacritty, Ghostty, iTerm2, WezTerm, and Microsoft Terminal all support it, as do editors like Neovim and Vim. With Warp now joining this list, TUI applications built on the protocol can leverage its full capabilities within Warp sessions. For everyday users, this means fewer misfired shortcuts and more reliable key combinations in tools like Neovim, terminal multiplexers, and interactive CLI utilities.

Customizable Agent Toolbelt

The agent input footer received a personalization upgrade: users can now drag and drop to rearrange context chips and controls in the toolbelt. Previously, the set and order of context items in the agent input bar — file references, directory chips, tool controls — was fixed. The new customizable toolbelt lets developers surface the context items they use most frequently at the front, and tuck away less-used ones.

This is a quality-of-life improvement particularly valuable for developers running frequent agent sessions with predictable context patterns — for example, always including a specific set of files or tools at the start of a prompt. Rather than manually re-adding context each session, the toolbelt now remembers the preferred layout.

Shell History Privacy Control

Warp added a new setting to hide agent-executed commands from shell history. Enabled by default, this prevents commands that Warp's AI agent runs on behalf of the user from appearing in ~/.zsh_history or ~/.bash_history. For users who prefer a clean, human-authored command history, this keeps AI-generated commands out of the historical record. The setting can be toggled off for users who prefer full transparency over what was executed in their session.

Additional Changes

Agent thinking blocks — the intermediate reasoning steps surfaced by certain AI models during complex tasks — can now be set to always show or always hide via Settings > AI > Other, rather than toggling them on a per-session basis. Dockerfile syntax highlighting was also added to Warp's native file editor.