SeaWorkSeaWork
English
Download

Orchestrate coding agents
from your desk and your phone

Run any coding agent from your phone, desktop, or terminal. Self-hosted and multi-provider.

All download optionsWeb AppGoogle Play
Run agents on a remote machine
npm install -g @seawork/cli && seawork
Supports
Claude CodeCodex
Self-hosted daemonE2E encrypted relayNative agent CLIs
mainseawork/website
Commit
+286-42
Orchestrator
Implement
Review
Polish the SeaWork website hero and mobile previews

I'll split this into design, copy, and implementation.

Run plan-designclaude visual direction
Run plan-technicalcodex component map
Wait for agentsplan-design plan-technical

Got the plans. Spinning up Codex for the refresh.

Run implementcodex 12 files changed

Implementation done. Asking Claude for a visual review.

Run reviewclaude spacing and copy checked

Homepage, docs, and downloads are ready.

Message the agent, tag @files, or use /commands and /skills
Opus 4.7
npm run dev
$ npm run dev --workspace=@seawork/website
> @seawork/website@1.0.7 dev
> vite --host 127.0.0.1
VITE v7.3.1 ready in 487ms
Local: http://127.0.0.1:8082/
hmr update /src/components/landing-hero.astro
hmr update /src/components/landing-phone-showcase.astro
route /download rebuilt in 132ms
route /docs rebuilt in 94ms
preview refreshed

When you want to step away from your desk, you can.

The native mobile app has full feature parity with desktop.

Use the best agent for the job

Run multiple providers from a single interface. SeaWork runs the native agent harness as you'd normally run it, with your skills, config and MCP servers intact.

Claude Code

Long-form reasoning and review

Codex

Fast edits inside real repos

Native CLIs stay intact: skills, config, auth, and MCP servers remain yours.

Your agents, every surface

Run agents on your laptop, a VM, or a dev server. Control them from any device with a direct connection or an E2E encrypted relay.

Desktop
Web
Mobile
CLI

E2E Encrypted Relay

or

Direct Connection

MacBook Pro
Hetzner VM
Dev server

Keyboard-first

Every action has a shortcut. Panels, splits, agents - all from the keyboard.

Command surface
Switch panels
⌘1-9
Split vertical
⌘D
Split horizontal
⌘ShiftD
Close panel
⌘W
New agent
⌘N
Command palette
⌘K

Local voice

Fully local voice stack. Speech-to-text and text-to-speech run entirely on your machine, nothing leaves your network.

Local speech pipelineoffline-first

Refactor the auth middleware to use the new session store, then run the test suite

I'll update the auth middleware to use SessionStore instead of the legacy cookie-based approach. Let me refactor the middleware and update the tests.

Fully scriptable

Everything you can do in the app, you can do from the terminal.

Full CLI reference

Run agents

Launch agents locally or on any remote host. The --worktree flag spins up an isolated git branch so you can run multiple agents on the same repo without conflicts.

seawork run --agent codex \
  --repo ~/code/seawork \
  --worktree feat/mobile-review \
  "Audit the mobile review flow and open a PR"

Loops

Have one agent do the work, another verify the result, and loop until it passes. Built-in, no shell scripting needed.

seawork loop run \
  --agent claude \
  --reviewer codex \
  --until "npm run typecheck" \
  "Refactor the websocket reconnect path"

Schedules

Run agents on a cron schedule. Automate recurring tasks like dependency updates, security audits, or report generation.

seawork schedule create weekly-security \
  --cron "0 9 * * 1" \
  --agent codex \
  "Review dependency updates and file issues"

FAQ

Clear answers for the parts that should stay boring: privacy, setup, and compatibility.

Is this free?
Yes. SeaWork is free. You need Claude Code or Codex installed with your own credentials. Voice is local-first by default and can optionally use OpenAI speech providers if you configure them.
Does my code leave my machine?
SeaWork doesn't send your code anywhere. Agents run locally and talk to their own APIs as they normally would. For remote access, you can use the optional end-to-end encrypted relay, connect directly over your local network, or use your own tunnel.
What agents does it support?
Claude Code and Codex. Each agent runs as its own process using its own CLI. SeaWork doesn't modify or wrap their behavior.
Do I need the desktop app?
No. You can run the daemon headless with npm install -g @seawork/cli && seawork and use the CLI, web app, or mobile app to connect. The desktop app just bundles the daemon with a UI.
How does voice work?
Voice runs locally on your device by default. You talk, the app transcribes and sends it to your agent as text. Optionally, you can configure OpenAI speech providers for higher-quality transcription and text-to-speech. See the voice docs.
Can I connect from outside my network?
Yes. You can use the hosted relay (end-to-end encrypted, SeaWork can't read your traffic), set up your own tunnel (Tailscale, Cloudflare Tunnel, etc.), or expose the daemon port directly. See configuration.
Do I need git or GitHub?
No. SeaWork works in any directory. Worktrees are optional and only relevant if you use git. You can run agents anywhere you'd normally work.
Can I get banned for using SeaWork?

We can't make promises on behalf of providers.

That said, SeaWork launches the official first-party CLIs (Claude Code and Codex) as subprocesses. It doesn't extract tokens or call inference APIs directly. From the provider's perspective, usage through SeaWork is indistinguishable from running the CLI yourself.

I've been using SeaWork with all providers for months without issue.

How do worktrees work?
When you launch an agent with the worktree option (from the app, desktop, or CLI), SeaWork creates a git worktree and runs the agent inside it. The agent works on an isolated branch without touching your main working directory. See the worktrees docs.