klaas vs Claude Code --teleport: External Tool vs Built-in Feature

When Anthropic released Claude Code version 2.1 in January 2026, the headline feature was teleport—the ability to move sessions between devices. Start a coding task on your desktop, check progress from your phone, resume on your laptop. For developers who’d been managing AI coding assistants through improvised solutions, it seemed like the answer.

But teleport isn’t the only approach to this problem. External tools like klaas provide similar capabilities through a different architecture, with different trade-offs. Understanding these differences helps you choose the right approach for your workflow.

What Claude Code --teleport Actually Does

The teleport feature enables session mobility between Claude Code’s web interface and terminal. The key word is “mobility”—sessions move from one environment to another rather than being shared simultaneously.

The typical workflow starts on the web. You open claude.ai, begin a coding conversation, perhaps ask Claude to analyse a codebase or plan a refactoring. Later, you want to continue in your terminal where you have your full development environment. Running claude --teleport presents a picker showing available web sessions. Select one, and Claude Code verifies you’re in the correct repository, checks out the relevant git branch, and loads the complete conversation history including any files Claude edited.

Background tasks extend this further. Prefix any prompt with & to run it as a background task that continues even when you close your terminal. The /tasks command shows running background tasks and lets you teleport into them. This enables a workflow where you kick off time-consuming tasks and check on them later from different devices.

The mobile story integrates with this. Anthropic’s iOS app lets you monitor Claude Code sessions, approve tool calls that require permission, and even start new tasks. Combined with background tasks, you can maintain oversight of AI coding work without being chained to your desk.

The Constraints of --teleport

Teleport’s tight integration comes with constraints that might or might not matter for your situation.

The most significant: teleport is one-directional. You can pull web sessions into your terminal with --teleport, but you cannot push terminal sessions to the web. If you start work in the terminal and later want to continue on your phone, you don’t have that option through teleport itself.

GitHub is required. Your repositories must have the Claude GitHub app installed, and you must authenticate with a GitHub account. GitLab, Bitbucket, and self-hosted git servers aren’t supported. For developers whose workflows centre on GitHub, this is fine. For those using other platforms, it’s a blocker.

The feature requires being on a paid plan—Pro at $20/month, Max at $100-200/month, or Team/Enterprise tiers. The free tier doesn’t include teleport capabilities.

Most fundamentally, teleport exists within Anthropic’s ecosystem. Your sessions, conversation history, and code context flow through Anthropic’s infrastructure. You’re trusting their security and privacy practices, with limited ability to verify claims independently.

What klaas Does Differently

klaas takes a different philosophical approach. Rather than integrating deeply with a specific AI assistant, it provides generic terminal session sharing that happens to work well with AI coding tools.

The architecture is straightforward: klaas wraps any command you specify in a pseudo-terminal (PTY), capturing all input and output. This data is encrypted on your machine using end-to-end encryption, then streamed to klaas servers. You access your sessions through a web dashboard at app.klaas.sh, viewing output in real-time and sending input remotely.

Because klaas wraps the PTY layer rather than integrating with specific applications, it works with any command-line tool. Claude Code, certainly, but also Cursor’s CLI, Gemini Code, or any other AI assistant that runs in a terminal. It also works with non-AI tools: you could use klaas to monitor a long-running build, watch a training job, or keep an eye on any process where you want remote visibility.

The client is open-source under the MIT license. This means you can read exactly what code runs on your machine, verify the encryption implementation, and confirm that data handling matches the stated privacy policy. The encryption happens in the client before transmission—klaas servers receive ciphertext they cannot decrypt.

Use Case: Mobile Monitoring

Consider the scenario: you’ve started Claude Code on a complex refactoring task that will take a while. You need to leave your desk but want to monitor progress and approve any permission requests Claude makes.

With Claude Code teleport, you’d use background tasks. You prefix your prompt with &, close your terminal, and monitor through the iOS app or web interface. When Claude needs approval for something, you get a notification and can respond from your phone. This works smoothly within Anthropic’s ecosystem, with deep integration into Claude Code’s permission system.

With klaas, you’d start your Claude Code session through klaas: klaas instead of claude. Your session streams to klaas’s servers, and you view it through their web dashboard from any device. When Claude Code asks for permission, you see the prompt in your browser and can send input remotely—typing y and enter just as you would locally.

The teleport approach is more polished for this specific use case. Approval requests appear as structured prompts in a native app rather than terminal output in a web viewer. Background task management is a first-class feature rather than an emergent capability.

The klaas approach is more general. The same setup works regardless of which AI assistant you’re using. If you switch from Claude Code to another tool tomorrow, your monitoring workflow doesn’t change. And you have cryptographic assurance, verifiable through open-source code, that the service cannot read your session content.

Use Case: Multi-Device Development

A different scenario: you’re working on a coding task, need to step away, and want to resume exactly where you left off on a different machine.

Teleport handles this through session mobility. Your conversation and context exist in Anthropic’s cloud, accessible from any authenticated device. Running claude --teleport on any machine pulls down the session state, checks out the right branch, and resumes the conversation.

klaas handles this differently. Your session continues running on the original machine, and you access it remotely from the second machine. You’re not resuming the session on new hardware—you’re controlling the original session from a different location. The practical difference: with teleport, you could close your laptop entirely. With klaas, the original machine needs to stay running (though it can be headless or asleep, as long as the session persists).

For developers who routinely work from multiple physical machines, teleport’s session mobility model fits better. For developers who have a primary development machine and want remote access to it, klaas’s approach makes more sense.

The Security and Privacy Question

Security-conscious developers will notice the fundamental architectural difference.

Teleport routes through Anthropic’s infrastructure by design. Your code context, conversation history, and commands flow through their servers as part of normal operation. Anthropic has security practices, privacy policies, and compliance certifications. You’re trusting that these are implemented as documented and that Anthropic won’t be compelled to access your data by governments or compromise it through security failures.

klaas’s end-to-end encryption means the servers handle only ciphertext. Even if klaas’s infrastructure were compromised, attackers would obtain encrypted data they cannot decrypt. The encryption keys exist only on your devices. This is verifiable through the open-source client—you can audit the implementation rather than trusting claims.

Neither approach is objectively superior; they represent different points on the trust spectrum. Developers working with sensitive code—proprietary algorithms, security-critical systems, or anything subject to regulatory requirements—might prefer klaas’s verifiable encryption. Developers prioritising convenience and willing to trust Anthropic might prefer teleport’s seamless integration.

Ecosystem Lock-in

Teleport ties you to Anthropic’s ecosystem in several ways. The feature requires GitHub specifically. It works only with Claude Code. Session history and context live in Anthropic’s infrastructure.

For developers committed to Claude Code and satisfied with GitHub, this lock-in provides value. Deep integration enables smooth workflows that generic tools can’t match. Background tasks, mobile approval, session mobility—these features exist because teleport knows about Claude Code’s internals.

klaas’s generic approach avoids lock-in at the cost of polish. It works with any CLI tool, any code hosting platform, any development workflow. But it doesn’t know that your application is Claude Code specifically, so it can’t provide Claude-specific features like structured approval prompts or conversation-aware session management.

The trade-off crystallises around a question: how confident are you that Claude Code will remain your primary AI coding assistant? If you’re certain, teleport’s integration advantages compound. If you’re hedging—perhaps expecting to evaluate alternatives as the field evolves—klaas’s tool-agnostic approach preserves flexibility.

What Each Tool Does Best

Teleport excels when you’re fully committed to Claude Code and GitHub. The background task system is genuinely useful for long-running operations. Mobile monitoring through a native app provides better UX than browser-based terminal viewing. Session mobility between devices works smoothly when you need it.

klaas excels when you want verifiable security, tool independence, or work with multiple AI assistants. The open-source client lets you audit exactly what happens to your data. The generic PTY wrapper works with anything that runs in a terminal. End-to-end encryption provides cryptographic rather than policy-based privacy guarantees.

Combining Both

Nothing prevents using both tools for different situations. Teleport for your primary Claude Code workflow when you want tight integration. klaas when you’re using other tools, working with particularly sensitive code, or need capabilities that teleport doesn’t provide.

The tools occupy different architectural layers and don’t conflict. Teleport manages session state within Claude Code. klaas manages session access at the terminal layer. You could even use klaas to wrap a Claude Code session if you wanted remote access with klaas’s encryption model rather than Anthropic’s infrastructure.

The AI coding assistant landscape continues evolving rapidly. New tools emerge, existing tools gain capabilities, and developer workflows shift accordingly. Choosing approaches that preserve flexibility—whether that’s teleport’s ecosystem integration or klaas’s tool independence—helps you adapt as the field develops.

Questions? Join our GitHub Discussions or reach out on 𝕏 @klaas_sh.