Reference

FAQ

Common questions about Pilothouse — supported agents, Tailscale, telemetry, how it compares to Claude Code Remote Control and herdr, and what's on the roadmap.

General

What is Pilothouse, in one sentence?

A single open-source binary that gives you one inbox for every coding agent on your machine, checks your rules before every tool call, and sends permission requests to your phone when you’re away.

Do I need to change how I start agents?

No. Pilothouse uses the agents’ official hooks, so claude and codex started from any terminal or IDE are covered. Wrap mode is optional.

Does it work inside tmux, herdr or over SSH?

Yes. Pilothouse hooks into the agent, not the terminal.

Which agents are supported?

Claude Code and Codex in v0.1. See the support table.

Is it free?

Yes. Pilothouse is Apache-2.0 and the open-source product is complete for individuals. A hosted relay and team features may become Pilothouse Cloud later.

Privacy and security

Does Pilothouse collect telemetry?

No. The only automatic network request is an anonymous daily update check (version, OS, architecture and a random ID that rotates monthly). Turn it off with update_check: false or PILOTHOUSE_NO_UPDATE_CHECK=1.

Does anything go through your servers?

No. There is no Pilothouse account and no relay in v0.1. Your phone talks to your machine over your tailnet.

Can an agent approve its own requests?

Pilothouse makes it hard: decisions only through the Unix socket with a process-ancestry check, signed device decisions for remote answers, and a built-in rule that asks before an agent touches Pilothouse. It’s defense in depth, not a sandbox — details in the security model.

Comparisons

How is it different from Claude Code’s Remote Control?

Remote Control is first-party, needs no install and is excellent if you only use Claude Code on a Claude subscription. Pilothouse is vendor-neutral (Claude Code and Codex in one inbox), adds your own rules before every tool call, works with API-key setups, and keeps everything on your machine.

How is it different from herdr?

herdr is a terminal multiplexer for agents. Pilothouse is a layer above whatever terminal you use, herdr included, focused on rules and approvals. They work well together. See Pilothouse vs herdr.

Why not just a Notification hook and ntfy?

That works well for alerts. Pilothouse adds answering from the phone with signed decisions, policies before tool calls, presence so you aren’t buzzed at your desk, and one view of every session.

Roadmap

These are not in v0.1. We’re saying so plainly so you can plan around it.

Planned What it means
End-to-end encrypted relay Answer from your phone without Tailscale. Open source and self-hostable; also offered hosted as part of Pilothouse Cloud.
Pilothouse Cloud Hosted relay, plus team features: shared policies, a team inbox, audit export. Waitlist only — not built yet.
More agents OpenCode, GitHub Copilot CLI, Gemini CLI.
Windows Observe mode first, then wrap mode.
Headless tasks Run tasks through an agent’s structured protocol (e.g. Codex app-server, ACP) with no terminal.
Several machines in one inbox Today each machine runs its own daemon and web app.

Have a strong opinion about the order? Open an issue.

Found a mistake? Open an issue.