Compare

Pilothouse vs Claude Code Remote Control

Remote Control is Claude Code’s own way to continue a session from the Claude apps. Pilothouse is a vendor-neutral daemon on your machine: one inbox for Claude Code, Codex, OpenCode and Grok Build, with your rules before every tool call and no account or relay.

Choose Claude Remote Control if…

  • You only use Claude Code, already pay for Claude and have the Claude app.
  • You want zero setup: one command in an open session.
  • You want the whole conversation on your phone — chat, attachments, model and effort, diffs — in a native app.
  • You want admin controls for an organisation.

Choose Pilothouse if…

  • You run Codex, OpenCode or Grok Build too, and want one inbox and one set of rules for all of them.
  • You want rules that run before every tool call — deny force-push, ask before deploys — on top of each agent’s own settings.
  • You use API keys or Bedrock, or don’t want a vendor account and relay between your phone and your machine.
  • You want usage and cost across Claude Code, Codex and OpenCode, with a daily budget.

What Claude Remote Control is

Remote Control is built into Claude Code. Run /remote-control (or start with --rc) and the session on your computer shows up in claude.ai/code and the Claude iOS and Android apps: the live transcript, tool activity, diffs, permission prompts and questions, and a message box. Since August 2026 a connected machine also appears as a device card you can start new sessions on. Traffic goes through Anthropic’s API over TLS, and it needs a paid claude.ai plan (Pro, Max, Team or Enterprise); API keys and Bedrock aren’t supported.

Side by side

Pilothouse compared with Claude Remote Control
FeaturePilothouseClaude Remote Control
What it isSingle Go binary: daemon, CLI, terminal inbox, phone web appBuilt into Claude Code; drive local sessions from the Claude apps and claude.ai
LicenseYes: Apache-2.0No: Proprietary
PriceFreeIncluded in Claude Pro, Max, Team, Enterprise (an Owner enables it on Team/Enterprise)
PlatformsmacOS, Linux (Windows on the roadmap)Host: CLI, Claude Desktop, VS Code. Phone: Claude iOS/Android apps, web
AgentsClaude Code, Codex, OpenCode, Grok BuildClaude Code only
Keep your terminal and how you launch agentsYes: Yes — official hooks (OpenCode: a plugin; Grok Build: personal hooks); any terminal, even inside tmux or herdrYes: Yes — /rc in a running session, or --rc
Your rules before every tool callYes: YAML + CEL, one-tap rules from approval cards, presets; any deny winsPartly: Claude Code’s own allow/ask/deny rules and permission modes
Approve from your phoneYes: Yes — each device signs its decisionsYes: Yes — prompts stay open until you answer
Answer questions / approve plans from your phoneYes: Claude Code questions and plans, OpenCode questions; Codex questions shown, answered in its terminalYes: Questions, yes; plan approval not spelled out in the docs
Reply to a running session from your phoneYes: Sessions started with pilothouse run, Claude Code sessions and OpenCodeYes: Yes — messages, photos and files
Live terminal from your phonePartly: For sessions started with pilothouse runNo: No — a conversation view, not a terminal
See what the session didYes: Session report with proof (tests, lint, build, files, commit), activity feed and diffs (Claude Code, Codex)Yes: Live transcript, tool activity, subagents, diffs
Token use and costYes: Tokens and cost at API prices (Claude Code, Codex, OpenCode) beside your plans, daily budget alertsPartly: /usage and /context as text; no budgets
Phone notificationsYes: Web Push, Telegram, ntfy (with buttons), Slack/Discord webhooks, desktopYes: Yes — when action is needed; Claude decides when to push
Presence-aware (no phone buzz while you’re at the desk)Yes: Yes — OS idle time; manual overrideYes: Skips push while you type in or focus the terminal
Answer from a chat appYes: Telegram, two-way: buttons, question options, repliesPartly: Separate Channels preview (Telegram, Discord, iMessage)
Native mobile appNo: No — an installable web app (PWA)Yes: Claude iOS and Android apps
Works without an accountYes: No accountNo: Needs a paid claude.ai plan; not API keys or Bedrock
How the phone reaches your machineYour own Tailscale (HTTPS via Tailscale Serve), no ports opened. Telegram needs no Tailscale.Through Anthropic’s API over TLS; the transcript is stored by Anthropic while connected
Worktree tasksYes: pilothouse task new → worktree + branch + agent; .pilothouse/task.yamlPartly: Server mode can give each session its own worktree
Start a task from your phoneYes: Yes — start it, review the diff, ship it as a PRYes: Yes — start a session on a connected machine
Several machines in one viewYes: Yes — link machines to one hub over your tailnetYes: Yes — each machine is a device card

Where Claude Remote Control is better

Native apps, first-party

The Claude iOS and Android apps with native push, supported by Anthropic. Pilothouse ships an installable web app that reaches your phone over Tailscale, plus a two-way Telegram channel.

Zero setup

One command in a running session and a QR code. Pilothouse needs a one-time pilothouse setup.

The full conversation

Subagent progress, a diff pane, slash commands like /model and /compact, photos and files. Pilothouse’s session page shows the activity feed and a reply box, not a full chat client.

Several machines with no setup

Every machine running Remote Control appears as a device card through your Claude account. Pilothouse links machines too, but each one needs Tailscale and a one-time `pilothouse link`.

Organisation controls

An Owner toggle, custom roles and Trusted Devices. Pilothouse has no team features (see Pilothouse Cloud).

Where Pilothouse is different

Cross-vendor

Claude Code, Codex, OpenCode and Grok Build sessions in one inbox, with the same rules, the same cards and the same phone.

Your rules before every tool call

YAML and CEL policies, one-tap rules from approval cards, and presets like guarded autopilot. Remote Control relies on Claude Code’s own permission rules and modes.

No account, no relay

The web app is served by your own machine over your tailnet; decisions from the web app are signed by a key on your phone. Nothing about your sessions passes through a vendor.

Cost across agents

Tokens and cost at API prices per day, agent, model and repo for Claude Code, Codex and OpenCode, beside what your plans cost, with alerts at 80 % and 100 % of a daily budget.

Using both

They don’t conflict. Pilothouse uses Claude Code’s hooks, so its rules still run on a session you drive through Remote Control, and whichever answer comes first — the Claude app, Pilothouse or the terminal — wins.

Facts about Claude Remote Control checked on 24 October 2026 from: Remote Control docs, Claude Code on mobile, What’s new, week 34, Permission modes. Spotted something out of date? Tell us.

Try Pilothouse alongside what you use today

It doesn't replace your terminal or your agents. Try the phone app with pretend agents first, or install, runpilothouse setup, and keep working.

Try it firstpilothouse demo
macOS · Linux
curl -fsSL https://pilothouse.dev/install.sh | sh

then run pilothouse setup