Warp Brings Its Full AI Coding Agent to Remote Servers Over SSH

If you spend any meaningful time SSHed into remote servers, you know the drill: you lose your IDE, your file explorer, and any smart tooling the moment you cross the network boundary. Warp just closed a big chunk of that gap. The terminal's SSH extension now delivers its full coding suite to remote macOS and Linux hosts , file tree, code review panel, and a fully capable AI coding agent , over a plain SSH connection.

What actually ships over the wire

Warp's SSH extension brings the local Warp experience to remote macOS and Linux hosts. After you opt in on first connect, you get a real file tree backed by the remote filesystem, Warp's code editor and code review on remote files, codebase indexing for the remote repo, and a coding agent that applies edits with Warp's native diff tool instead of falling back to sed.

Here's the full list of what's now available over SSH:

• File tree (Project Explorer) , the file tree sidebar reflects the remote project structure and updates as you navigate or change files.
• Codebase indexing , Warp indexes the remote repository so Agents can semantically search the remote codebase, just like Codebase Context locally.
• Native code diffs , the Agent reads files and applies edits through Warp's built-in diff tool. Code changes show up as inline diffs you can review and approve, instead of being applied via sed or other shell commands.
• Project rules and skills , Agents discover and use the remote project's rules (WARP.md) and skills, the same as in a local repo.
• Reliable completions , generators run in parallel over a single multiplexed connection instead of opening a new SSH session per command, so completions stop hitting the remote host's MaxSessions ceiling.
• File search , search for remote files by name from the Command Palette.

The clever engineering underneath

When you connect to a remote host and accept the install prompt, Warp installs a small companion server in your home directory on the remote machine. The server runs in the background as your user and communicates with Warp over your existing SSH connection. It doesn't open any network ports, doesn't require root, and doesn't change anything outside your home directory. The server does the heavy lifting directly on the remote machine: it reads and writes files, watches your repository for changes, runs git commands, and indexes your codebase.

The key insight here is where the work happens. Only lightweight, incremental updates stream back to Warp , an edit, a diff, a status change , instead of full file contents round-tripped through shell commands. This keeps the connection fast while features like the file tree, code review, and the Agent's file tools behave the same as they do locally. It's a fundamentally different architecture from tunneling a full protocol like LSP over SSH.

A few practical properties worth knowing:

• Multiple SSH sessions and Warp windows connected to the same host share a single server process.
• The extension version tracks your Warp version. When you connect to a host with an outdated extension, Warp automatically installs the matching version.
• Uninstalling is as simple as deleting the ~/.warp*/remote-server directory on the remote host.

What's still missing

It's not a full local replacement yet. Warp's native code editor UI is not yet available in remote sessions , you cannot open, view, or edit remote files in Warp's built-in editor. The Agent also cannot take screenshots or interact with the remote machine's desktop. And WSL sessions have the same limitations as SSH and are not currently covered by the SSH extension.

Getting it running

The setup is intentionally low-friction. On your next SSH connection to a remote host, Warp will prompt you to install the extension. Accept it, and the companion server is downloaded and started automatically. If you want to skip the prompt entirely, you can set Settings > Warpify to always install silently. For hosts you never want to Warpify, there's a denylist.

Requirements on the remote side:

• Linux or macOS (Windows remote hosts are not supported)
• x86_64 or aarch64 architecture
• glibc 2.31+ on Linux (Ubuntu 20.04 equivalent , older distros like RHEL 7/8, Amazon Linux 2, and Alpine are not supported)
• bash or zsh as the remote shell
• Outbound HTTPS to app.warp.dev for the initial install (or Warp uploads the binary over the SSH connection if curl/wget aren't available)

Where this fits in the bigger picture

Warp in 2026 is a meaningfully different product than the macOS-only AI terminal that launched in 2022. It's GA on every major desktop OS, the client is open source under MIT + AGPL v3, and the AI surface has split into Active AI (always-on hints), Agent Mode (interactive multi-step), and Cloud Agents on Oz (autonomous, event-triggered). The SSH extension is the piece that makes all of that usable on the machines where a lot of real work actually runs.

The practical use cases are significant: debugging production servers without switching to a separate editor, running the AI agent against a staging environment's actual codebase, or doing code review on a remote repo without cloning it locally first. Code indexed with Codebase Context is never stored on Warp's servers, which matters when the remote machine holds sensitive code.

The SSH extension is available now on all plans. Warp has a free tier with $0 monthly cost, offering limited AI credits and basic terminal features. The Build plan is $20/month with 1,500 monthly credits, plus BYOK for OpenAI, Anthropic, or Google. See the official SSH extension docs and the SSH feature support matrix for the full breakdown of what works and what doesn't.