Redock vs KittyLitter

KittyLitter—whose underlying project is also referred to as Litter—is designed first as a native mobile client for Codex. Its interface renders the conversation, reasoning, code blocks, images, and session history in mobile-native components. Local discovery and SSH can start or reach an agent on another machine, while the app keeps the user focused on the agent rather than on a shell.

Redock starts one layer lower. It is a full remote terminal and then adds an AI coding workflow above it. Claude Code, Codex, OpenCode, or another terminal agent runs in the same environment where the developer can execute ordinary commands, enter tmux, inspect a service, use a TUI, or launch a named Project Action.

The distinction is not whether both products can help an agent write code. They can. The distinction is what happens when the task leaves the agent conversation. KittyLitter keeps the interaction intentionally focused; Redock keeps the rest of the development machine available without changing apps.

KittyLitter’s clearest advantage is price and openness. The app is free, the project is open source under GPLv3, and in-app purchases are optional tips rather than a required feature unlock. For developers who want to inspect the client that reaches their coding environment, this is a meaningful product decision rather than a marketing detail.

Codex is the center of the experience. The App Store description emphasizes local and remote Codex servers, mobile session management, conversation history, reasoning, code blocks, images, and voice input. Recent versions also describe beta support for Claude Code, OpenCode, and Pi, but Codex remains the most established path.

Running Codex on the device is the standout capability. A user can work without keeping a separate development server available for every session. That model does not replace a full workstation for large repositories or heavy builds, but it offers a kind of independence Redock does not: Redock always connects to a development host where the terminal and agent run.

KittyLitter can find compatible Codex servers on the local network automatically. That is the easiest path at home or in an office: discover the machine, choose it, and begin without manually typing an address. For a conventional remote server, the app can connect over SSH and start Codex remotely.

Its more unusual option is a peer-to-peer path designed to cross NAT and firewall boundaries without a conventional hosted session relay. Moshi’s reference analysis describes the companion tunnel as alleycat, built on Iroh and QUIC with QR pairing. That approach is attractive when installing a mesh VPN or forwarding a port would otherwise be the largest setup hurdle.

Redock uses familiar terminal connectivity. SSH provides the universal baseline, Mosh improves roaming and recovery across network changes, and users can place the host behind Tailscale or another VPN when direct exposure is undesirable. KittyLitter has the more integrated peer-to-peer story; Redock has the more conventional and interoperable terminal transport.

KittyLitter is an agent client, not a general terminal emulator. That focus produces a readable conversation interface, but it means the user cannot treat the session as an arbitrary PTY. When the next step is tailing a log, running a database console, opening Vim, using htop, managing another host, or executing a command the agent does not expose, a separate terminal is still necessary.

Redock keeps the coding agent inside a full shell. The developer can pause or interrupt the agent, run an exact verification command, inspect a process, use any installed TUI, and then return to the same session. This is especially useful when AI coding and ordinary server work are mixed rather than neatly separated.

A full terminal also changes how the product handles failure. If an agent integration breaks, its renderer does not understand a new output type, or a beta agent path becomes unstable, Redock can fall back to the raw command-line program. KittyLitter can create a much better specialized experience when the integration is working, but its boundary is narrower by design.

KittyLitter preserves and renders the coding-agent session. That is the right unit of continuity when the conversation is all the user needs. The app can reconnect, display prior messages, and keep the mobile presentation independent from a raw terminal buffer.

Redock puts continuity in tmux on the development host. The same agent process can be started from a laptop, attached from Redock, and later resumed from any ordinary terminal. The continuity includes everything in the pane—not just agent messages, but the shell state, commands, interactive programs, and scrollback around them.

Mosh adds a second layer for mobile use by making the network connection tolerant of roaming and temporary loss. It does not keep the process alive by itself, which is why tmux remains important. Redock’s model is more general; KittyLitter’s model can be more legible because it understands the agent conversation directly.

Both products recognize that typing a long coding requirement on a phone is unpleasant. KittyLitter offers voice input directly in its agent-native experience, and its realtime, conversational feel can be the fastest way to keep Codex moving. That voice path is included in a free app, which is a genuine advantage.

Redock uses voice in a more staged model. A developer can dictate a requirement, review and edit it, then send the complete text to the active terminal agent. The surrounding Project remembers the host and repository directory, while named Actions handle deterministic tasks that do not need another natural-language round trip.

For example, an agent can implement a feature from a dictated requirement, but the canonical tests and staging deployment can remain exact Actions. Their outputs become Runs and recent work appears in Activity. KittyLitter keeps the experience centered on what the agent says and does; Redock adds a workflow for the commands and outcomes around that conversation.

KittyLitter is a young, fast-moving product. That creates excitement and risk at the same time. The Codex path is the strongest, while support for several other agents is described as beta. Features and connection methods can change quickly, and some workflows will have rougher edges than a long-established terminal client.

The upside is equally real: a small open-source product can move quickly, expose novel connectivity, add agent support, and let users inspect or contribute to the implementation. Developers who enjoy early tools and primarily use Codex may value that momentum more than polish or workflow breadth.

Choose Redock when the mobile tool must remain useful even outside the agent layer and you prefer a settled Project, Action, Run, Activity, SSH/Mosh, and tmux model. Choose KittyLitter when free and open source are non-negotiable, Codex is the main agent, on-device use is appealing, and a focused agent UI is more important than a general shell.